address comments

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

AGENTS.md

@@ -167,284 +167,7 @@ web/
 
 ## Frontend Standards
 
-### 1. Import Standards
-
-**Always use absolute imports with the `@` prefix.**
-
-**Reason:** Moving files around becomes easier since you don't also have to update those import statements. This makes modifications to the codebase much nicer.
-
-```typescript
-// ✅ Good
-import { Button } from "@/components/ui/button";
-import { useAuth } from "@/hooks/useAuth";
-import { Text } from "@/refresh-components/texts/Text";
-
-// ❌ Bad
-import { Button } from "../../../components/ui/button";
-import { useAuth } from "./hooks/useAuth";
-```
-
-### 2. React Component Functions
-
-**Prefer regular functions over arrow functions for React components.**
-
-**Reason:** Functions just become easier to read.
-
-```typescript
-// ✅ Good
-function UserProfile({ userId }: UserProfileProps) {
-  return <div>User Profile</div>
-}
-
-// ❌ Bad
-const UserProfile = ({ userId }: UserProfileProps) => {
-  return <div>User Profile</div>
-}
-```
-
-### 3. Props Interface Extraction
-
-**Extract prop types into their own interface definitions.**
-
-**Reason:** Functions just become easier to read.
-
-```typescript
-// ✅ Good
-interface UserCardProps {
-  user: User
-  showActions?: boolean
-  onEdit?: (userId: string) => void
-}
-
-function UserCard({ user, showActions = false, onEdit }: UserCardProps) {
-  return <div>User Card</div>
-}
-
-// ❌ Bad
-function UserCard({
-  user,
-  showActions = false,
-  onEdit
-}: {
-  user: User
-  showActions?: boolean
-  onEdit?: (userId: string) => void
-}) {
-  return <div>User Card</div>
-}
-```
-
-### 4. Spacing Guidelines
-
-**Prefer padding over margins for spacing.**
-
-**Reason:** We want to consolidate usage to paddings instead of margins.
-
-```typescript
-// ✅ Good
-<div className="p-4 space-y-2">
-  <div className="p-2">Content</div>
-</div>
-
-// ❌ Bad
-<div className="m-4 space-y-2">
-  <div className="m-2">Content</div>
-</div>
-```
-
-### 5. Tailwind Dark Mode
-
-**Strictly forbid using the `dark:` modifier in Tailwind classes, except for logo icon handling.**
-
-**Reason:** The `colors.css` file already, VERY CAREFULLY, defines what the exact opposite colour of each light-mode colour is. Overriding this behaviour is VERY bad and will lead to horrible UI breakages.
-
-**Exception:** The `createLogoIcon` helper in `web/src/components/icons/icons.tsx` uses `dark:` modifiers (`dark:invert`, `dark:hidden`, `dark:block`) to handle third-party logo icons that cannot automatically adapt through `colors.css`. This is the ONLY acceptable use of dark mode modifiers.
-
-```typescript
-// ✅ Good - Standard components use `tailwind-themes/tailwind.config.js` / `src/app/css/colors.css`
-<div className="bg-background-neutral-03 text-text-02">
-  Content
-</div>
-
-// ✅ Good - Logo icons with dark mode handling via createLogoIcon
-export const GithubIcon = createLogoIcon(githubLightIcon, {
-  monochromatic: true,  // Will apply dark:invert internally
-});
-
-export const GitbookIcon = createLogoIcon(gitbookLightIcon, {
-  darkSrc: gitbookDarkIcon,  // Will use dark:hidden/dark:block internally
-});
-
-// ❌ Bad - Manual dark mode overrides
-<div className="bg-white dark:bg-black text-black dark:text-white">
-  Content
-</div>
-```
-
-### 6. Class Name Utilities
-
-**Use the `cn` utility instead of raw string formatting for classNames.**
-
-**Reason:** `cn`s are easier to read. They also allow for more complex types (i.e., string-arrays) to get formatted properly (it flattens each element in that string array down). As a result, it can allow things such as conditionals (i.e., `myCondition && "some-tailwind-class"`, which evaluates to `false` when `myCondition` is `false`) to get filtered out.
-
-```typescript
-import { cn } from '@/lib/utils'
-
-// ✅ Good
-<div className={cn(
-  'base-class',
-  isActive && 'active-class',
-  className
-)}>
-  Content
-</div>
-
-// ❌ Bad
-<div className={`base-class ${isActive ? 'active-class' : ''} ${className}`}>
-  Content
-</div>
-```
-
-### 7. Custom Hooks Organization
-
-**Follow a "hook-per-file" layout. Each hook should live in its own file within `web/src/hooks`.**
-
-**Reason:** This is just a layout preference. Keeps code clean.
-
-```typescript
-// web/src/hooks/useUserData.ts
-export function useUserData(userId: string) {
-  // hook implementation
-}
-
-// web/src/hooks/useLocalStorage.ts
-export function useLocalStorage<T>(key: string, initialValue: T) {
-  // hook implementation
-}
-```
-
-### 8. Icon Usage
-
-**ONLY use icons from the `web/src/icons` directory. Do NOT use icons from `react-icons`, `lucide`, or other external libraries.**
-
-**Reason:** We have a very carefully curated selection of icons that match our Onyx guidelines. We do NOT want to muddy those up with different aesthetic stylings.
-
-```typescript
-// ✅ Good
-import SvgX from "@/icons/x";
-import SvgMoreHorizontal from "@/icons/more-horizontal";
-
-// ❌ Bad
-import { User } from "lucide-react";
-import { FiSearch } from "react-icons/fi";
-```
-
-**Missing Icons**: If an icon is needed but doesn't exist in the `web/src/icons` directory, import it from Figma using the Figma MCP tool and add it to the icons directory.
-If you need help with this step, reach out to `raunak@onyx.app`.
-
-### 9. Text Rendering
-
-**Prefer using the `refresh-components/texts/Text` component for all text rendering. Avoid "naked" text nodes.**
-
-**Reason:** The `Text` component is fully compliant with the stylings provided in Figma. It provides easy utilities to specify the text-colour and font-size in the form of flags. Super duper easy.
-
-```typescript
-// ✅ Good
-import { Text } from '@/refresh-components/texts/Text'
-
-function UserCard({ name }: { name: string }) {
-  return (
-    <Text
-      {/* The `text03` flag makes the text it renders to be coloured the 3rd-scale grey */}
-      text03
-      {/* The `mainAction` flag makes the text it renders to be "main-action" font + line-height + weightage, as described in the Figma */}
-      mainAction
-    >
-      {name}
-    </Text>
-  )
-}
-
-// ❌ Bad
-function UserCard({ name }: { name: string }) {
-  return (
-    <div>
-      <h2>{name}</h2>
-      <p>User details</p>
-    </div>
-  )
-}
-```
-
-### 10. Component Usage
-
-**Heavily avoid raw HTML input components. Always use components from the `web/src/refresh-components` or `web/lib/opal/src` directory.**
-
-**Reason:** We've put in a lot of effort to unify the components that are rendered in the Onyx app. Using raw components breaks the entire UI of the application, and leaves it in a muddier state than before.
-
-```typescript
-// ✅ Good
-import Button from '@/refresh-components/buttons/Button'
-import InputTypeIn from '@/refresh-components/inputs/InputTypeIn'
-import SvgPlusCircle from '@/icons/plus-circle'
-
-function ContactForm() {
-  return (
-    <form>
-      <InputTypeIn placeholder="Search..." />
-      <Button type="submit" leftIcon={SvgPlusCircle}>Submit</Button>
-    </form>
-  )
-}
-
-// ❌ Bad
-function ContactForm() {
-  return (
-    <form>
-      <input placeholder="Name" />
-      <textarea placeholder="Message" />
-      <button type="submit">Submit</button>
-    </form>
-  )
-}
-```
-
-### 11. Colors
-
-**Always use custom overrides for colors and borders rather than built in Tailwind CSS colors. These overrides live in `web/tailwind-themes/tailwind.config.js`.**
-
-**Reason:** Our custom color system uses CSS variables that automatically handle dark mode and maintain design consistency across the app. Standard Tailwind colors bypass this system.
-
-**Available color categories:**
-
-- **Text:** `text-01` through `text-05`, `text-inverted-XX`
-- **Backgrounds:** `background-neutral-XX`, `background-tint-XX` (and inverted variants)
-- **Borders:** `border-01` through `border-05`, `border-inverted-XX`
-- **Actions:** `action-link-XX`, `action-danger-XX`
-- **Status:** `status-info-XX`, `status-success-XX`, `status-warning-XX`, `status-error-XX`
-- **Theme:** `theme-primary-XX`, `theme-red-XX`, `theme-blue-XX`, etc.
-
-```typescript
-// ✅ Good - Use custom Onyx color classes
-<div className="bg-background-neutral-01 border border-border-02" />
-<div className="bg-background-tint-02 border border-border-01" />
-<div className="bg-status-success-01" />
-<div className="bg-action-link-01" />
-<div className="bg-theme-primary-05" />
-
-// ❌ Bad - Do NOT use standard Tailwind colors
-<div className="bg-gray-100 border border-gray-300 text-gray-600" />
-<div className="bg-white border border-slate-200" />
-<div className="bg-green-100 text-green-700" />
-<div className="bg-blue-100 text-blue-600" />
-<div className="bg-indigo-500" />
-```
-
-### 12. Data Fetching
-
-**Prefer using `useSWR` for data fetching. Data should generally be fetched on the client side. Components that need data should display a loader / placeholder while waiting for that data. Prefer loading data within the component that needs it rather than at the top level and passing it down.**
-
-**Reason:** Client side fetching allows us to load the skeleton of the page without waiting for data to load, leading to a snappier UX. Loading data where needed reduces dependencies between a component and its parent component(s).
+Frontend standards for the `web/` and `desktop/` projects live in `web/AGENTS.md`.
 
 ## Database & Migrations
 

backend/Dockerfile

@@ -47,6 +47,8 @@ RUN apt-get update && \
         gcc \
         nano \
         vim \
+        # Install procps so kubernetes exec sessions can use ps aux for debugging
+        procps \
         libjemalloc2 \
         && \
     rm -rf /var/lib/apt/lists/* && \

backend/onyx/background/celery/apps/primary.py

@@ -317,6 +317,7 @@ celery_app.autodiscover_tasks(
             "onyx.background.celery.tasks.docprocessing",
             "onyx.background.celery.tasks.evals",
             "onyx.background.celery.tasks.hierarchyfetching",
+            "onyx.background.celery.tasks.hooks",
             "onyx.background.celery.tasks.periodic",
             "onyx.background.celery.tasks.pruning",
             "onyx.background.celery.tasks.shared",

backend/onyx/background/celery/tasks/beat_schedule.py

@@ -14,6 +14,7 @@ from onyx.configs.constants import ONYX_CLOUD_CELERY_TASK_PREFIX
 from onyx.configs.constants import OnyxCeleryPriority
 from onyx.configs.constants import OnyxCeleryQueues
 from onyx.configs.constants import OnyxCeleryTask
+from onyx.hooks.utils import HOOKS_AVAILABLE
 from shared_configs.configs import MULTI_TENANT
 
 # choosing 15 minutes because it roughly gives us enough time to process many tasks
@@ -361,6 +362,19 @@ if not MULTI_TENANT:
 
     tasks_to_schedule.extend(beat_task_templates)
 
+if HOOKS_AVAILABLE:
+    tasks_to_schedule.append(
+        {
+            "name": "hook-execution-log-cleanup",
+            "task": OnyxCeleryTask.HOOK_EXECUTION_LOG_CLEANUP_TASK,
+            "schedule": timedelta(days=1),
+            "options": {
+                "priority": OnyxCeleryPriority.LOW,
+                "expires": BEAT_EXPIRES_DEFAULT,
+            },
+        }
+    )
+
 
 def generate_cloud_tasks(
     beat_tasks: list[dict], beat_templates: list[dict], beat_multiplier: float

backend/onyx/background/celery/tasks/hooks/tasks.py

@@ -0,0 +1,35 @@
+from celery import shared_task
+
+from onyx.configs.app_configs import JOB_TIMEOUT
+from onyx.configs.constants import OnyxCeleryTask
+from onyx.db.engine.sql_engine import get_session_with_current_tenant
+from onyx.db.hook import cleanup_old_execution_logs__no_commit
+from onyx.utils.logger import setup_logger
+
+logger = setup_logger()
+
+_HOOK_EXECUTION_LOG_RETENTION_DAYS: int = 30
+
+
+@shared_task(
+    name=OnyxCeleryTask.HOOK_EXECUTION_LOG_CLEANUP_TASK,
+    ignore_result=True,
+    soft_time_limit=JOB_TIMEOUT,
+    trail=False,
+)
+def hook_execution_log_cleanup_task(*, tenant_id: str) -> None:  # noqa: ARG001
+    try:
+        with get_session_with_current_tenant() as db_session:
+            deleted: int = cleanup_old_execution_logs__no_commit(
+                db_session=db_session,
+                max_age_days=_HOOK_EXECUTION_LOG_RETENTION_DAYS,
+            )
+            db_session.commit()
+            if deleted:
+                logger.info(
+                    f"Deleted {deleted} hook execution log(s) older than "
+                    f"{_HOOK_EXECUTION_LOG_RETENTION_DAYS} days."
+                )
+    except Exception:
+        logger.exception("Failed to clean up hook execution logs")
+        raise

backend/onyx/background/celery/tasks/user_file_processing/tasks.py

@@ -24,6 +24,7 @@ from onyx.configs.app_configs import MANAGED_VESPA
 from onyx.configs.app_configs import VESPA_CLOUD_CERT_PATH
 from onyx.configs.app_configs import VESPA_CLOUD_KEY_PATH
 from onyx.configs.constants import CELERY_GENERIC_BEAT_LOCK_TIMEOUT
+from onyx.configs.constants import CELERY_USER_FILE_DELETE_TASK_EXPIRES
 from onyx.configs.constants import CELERY_USER_FILE_PROCESSING_LOCK_TIMEOUT
 from onyx.configs.constants import CELERY_USER_FILE_PROCESSING_TASK_EXPIRES
 from onyx.configs.constants import CELERY_USER_FILE_PROJECT_SYNC_LOCK_TIMEOUT
@@ -33,6 +34,7 @@ from onyx.configs.constants import OnyxCeleryPriority
 from onyx.configs.constants import OnyxCeleryQueues
 from onyx.configs.constants import OnyxCeleryTask
 from onyx.configs.constants import OnyxRedisLocks
+from onyx.configs.constants import USER_FILE_DELETE_MAX_QUEUE_DEPTH
 from onyx.configs.constants import USER_FILE_PROCESSING_MAX_QUEUE_DEPTH
 from onyx.configs.constants import USER_FILE_PROJECT_SYNC_MAX_QUEUE_DEPTH
 from onyx.connectors.file.connector import LocalFileConnector
@@ -91,6 +93,17 @@ def _user_file_delete_lock_key(user_file_id: str | UUID) -> str:
     return f"{OnyxRedisLocks.USER_FILE_DELETE_LOCK_PREFIX}:{user_file_id}"
 
 
+def _user_file_delete_queued_key(user_file_id: str | UUID) -> str:
+    """Key that exists while a delete_single_user_file task is sitting in the queue.
+
+    The beat generator sets this with a TTL equal to CELERY_USER_FILE_DELETE_TASK_EXPIRES
+    before enqueuing and the worker deletes it as its first action.  This prevents
+    the beat from adding duplicate tasks for files that already have a live task
+    in flight.
+    """
+    return f"{OnyxRedisLocks.USER_FILE_DELETE_QUEUED_PREFIX}:{user_file_id}"
+
+
 def get_user_file_project_sync_queue_depth(celery_app: Celery) -> int:
     redis_celery: Redis = celery_app.broker_connection().channel().client  # type: ignore
     return celery_get_queue_length(
@@ -546,7 +559,23 @@ def process_single_user_file(
     ignore_result=True,
 )
 def check_for_user_file_delete(self: Task, *, tenant_id: str) -> None:
-    """Scan for user files with DELETING status and enqueue per-file tasks."""
+    """Scan for user files with DELETING status and enqueue per-file tasks.
+
+    Three mechanisms prevent queue runaway (mirrors check_user_file_processing):
+
+    1. **Queue depth backpressure** – if the broker queue already has more than
+       USER_FILE_DELETE_MAX_QUEUE_DEPTH items we skip this beat cycle entirely.
+
+    2. **Per-file queued guard** – before enqueuing a task we set a short-lived
+       Redis key (TTL = CELERY_USER_FILE_DELETE_TASK_EXPIRES).  If that key
+       already exists the file already has a live task in the queue, so we skip
+       it.  The worker deletes the key the moment it picks up the task so the
+       next beat cycle can re-enqueue if the file is still DELETING.
+
+    3. **Task expiry** – every enqueued task carries an `expires` value equal to
+       CELERY_USER_FILE_DELETE_TASK_EXPIRES.  If a task is still sitting in
+       the queue after that deadline, Celery discards it without touching the DB.
+    """
     task_logger.info("check_for_user_file_delete - Starting")
     redis_client = get_redis_client(tenant_id=tenant_id)
     lock: RedisLock = redis_client.lock(
@@ -555,8 +584,23 @@ def check_for_user_file_delete(self: Task, *, tenant_id: str) -> None:
     )
     if not lock.acquire(blocking=False):
         return None
+
     enqueued = 0
+    skipped_guard = 0
     try:
+        # --- Protection 1: queue depth backpressure ---
+        # NOTE: must use the broker's Redis client (not redis_client) because
+        # Celery queues live on a separate Redis DB with CELERY_SEPARATOR keys.
+        r_celery: Redis = self.app.broker_connection().channel().client  # type: ignore
+        queue_len = celery_get_queue_length(OnyxCeleryQueues.USER_FILE_DELETE, r_celery)
+        if queue_len > USER_FILE_DELETE_MAX_QUEUE_DEPTH:
+            task_logger.warning(
+                f"check_for_user_file_delete - Queue depth {queue_len} exceeds "
+                f"{USER_FILE_DELETE_MAX_QUEUE_DEPTH}, skipping enqueue for "
+                f"tenant={tenant_id}"
+            )
+            return None
+
         with get_session_with_current_tenant() as db_session:
             user_file_ids = (
                 db_session.execute(
@@ -568,23 +612,40 @@ def check_for_user_file_delete(self: Task, *, tenant_id: str) -> None:
                 .all()
             )
             for user_file_id in user_file_ids:
-                self.app.send_task(
-                    OnyxCeleryTask.DELETE_SINGLE_USER_FILE,
-                    kwargs={"user_file_id": str(user_file_id), "tenant_id": tenant_id},
-                    queue=OnyxCeleryQueues.USER_FILE_DELETE,
-                    priority=OnyxCeleryPriority.HIGH,
+                # --- Protection 2: per-file queued guard ---
+                queued_key = _user_file_delete_queued_key(user_file_id)
+                guard_set = redis_client.set(
+                    queued_key,
+                    1,
+                    ex=CELERY_USER_FILE_DELETE_TASK_EXPIRES,
+                    nx=True,
                 )
+                if not guard_set:
+                    skipped_guard += 1
+                    continue
+
+                # --- Protection 3: task expiry ---
+                try:
+                    self.app.send_task(
+                        OnyxCeleryTask.DELETE_SINGLE_USER_FILE,
+                        kwargs={
+                            "user_file_id": str(user_file_id),
+                            "tenant_id": tenant_id,
+                        },
+                        queue=OnyxCeleryQueues.USER_FILE_DELETE,
+                        priority=OnyxCeleryPriority.HIGH,
+                        expires=CELERY_USER_FILE_DELETE_TASK_EXPIRES,
+                    )
+                except Exception:
+                    redis_client.delete(queued_key)
+                    raise
                 enqueued += 1
-    except Exception as e:
-        task_logger.exception(
-            f"check_for_user_file_delete - Error enqueuing deletes - {e.__class__.__name__}"
-        )
-        return None
     finally:
         if lock.owned():
             lock.release()
+
     task_logger.info(
-        f"check_for_user_file_delete - Enqueued {enqueued} tasks for tenant={tenant_id}"
+        f"check_for_user_file_delete - Enqueued {enqueued} tasks, skipped_guard={skipped_guard} for tenant={tenant_id}"
     )
     return None
 
@@ -602,6 +663,9 @@ def delete_user_file_impl(
     file_lock: RedisLock | None = None
     if redis_locking:
         redis_client = get_redis_client(tenant_id=tenant_id)
+        # Clear the queued guard so the beat can re-enqueue if deletion fails
+        # and the file remains in DELETING status.
+        redis_client.delete(_user_file_delete_queued_key(user_file_id))
         file_lock = redis_client.lock(
             _user_file_delete_lock_key(user_file_id),
             timeout=CELERY_GENERIC_BEAT_LOCK_TIMEOUT,

backend/onyx/cache/postgres_backend.py

@@ -297,7 +297,9 @@ class PostgresCacheBackend(CacheBackend):
 
     def _lock_id_for(self, name: str) -> int:
         """Map *name* to a 64-bit signed int for ``pg_advisory_lock``."""
-        h = hashlib.md5(f"{self._tenant_id}:{name}".encode()).digest()
+        h = hashlib.md5(
+            f"{self._tenant_id}:{name}".encode(), usedforsecurity=False
+        ).digest()
         return struct.unpack("q", h[:8])[0]
 
 

backend/onyx/chat/llm_loop.py

@@ -36,9 +36,11 @@ from onyx.db.memory import add_memory
 from onyx.db.memory import update_memory_at_index
 from onyx.db.memory import UserMemoryContext
 from onyx.db.models import Persona
+from onyx.llm.constants import LlmProviderNames
 from onyx.llm.interfaces import LLM
 from onyx.llm.interfaces import LLMUserIdentity
 from onyx.llm.interfaces import ToolChoiceOptions
+from onyx.llm.utils import is_true_openai_model
 from onyx.prompts.chat_prompts import IMAGE_GEN_REMINDER
 from onyx.prompts.chat_prompts import OPEN_URL_REMINDER
 from onyx.server.query_and_chat.placement import Placement
@@ -72,6 +74,70 @@ from shared_configs.contextvars import get_current_tenant_id
 logger = setup_logger()
 
 
+class EmptyLLMResponseError(RuntimeError):
+    """Raised when the streamed LLM response completes without a usable answer."""
+
+    def __init__(
+        self,
+        *,
+        provider: str,
+        model: str,
+        tool_choice: ToolChoiceOptions,
+        client_error_msg: str,
+        error_code: str = "EMPTY_LLM_RESPONSE",
+        is_retryable: bool = True,
+    ) -> None:
+        super().__init__(client_error_msg)
+        self.provider = provider
+        self.model = model
+        self.tool_choice = tool_choice
+        self.client_error_msg = client_error_msg
+        self.error_code = error_code
+        self.is_retryable = is_retryable
+
+
+def _build_empty_llm_response_error(
+    llm: LLM,
+    llm_step_result: LlmStepResult,
+    tool_choice: ToolChoiceOptions,
+) -> EmptyLLMResponseError:
+    provider = llm.config.model_provider
+    model = llm.config.model_name
+
+    # OpenAI quota exhaustion has reached us as a streamed "stop" with zero content.
+    # When the stream is completely empty and there is no reasoning/tool output, surface
+    # the likely account-level cause instead of a generic tool-calling error.
+    if (
+        not llm_step_result.reasoning
+        and provider == LlmProviderNames.OPENAI
+        and is_true_openai_model(provider, model)
+    ):
+        return EmptyLLMResponseError(
+            provider=provider,
+            model=model,
+            tool_choice=tool_choice,
+            client_error_msg=(
+                "The selected OpenAI model returned an empty streamed response "
+                "before producing any tokens. This commonly happens when the API "
+                "key or project has no remaining quota or billing is not enabled. "
+                "Verify quota and billing for this key and try again."
+            ),
+            error_code="BUDGET_EXCEEDED",
+            is_retryable=False,
+        )
+
+    return EmptyLLMResponseError(
+        provider=provider,
+        model=model,
+        tool_choice=tool_choice,
+        client_error_msg=(
+            "The selected model returned no final answer before the stream "
+            "completed. No text or tool calls were received from the upstream "
+            "provider."
+        ),
+    )
+
+
 def _looks_like_xml_tool_call_payload(text: str | None) -> bool:
     """Detect XML-style marshaled tool calls emitted as plain text."""
     if not text:
@@ -613,7 +679,12 @@ def run_llm_loop(
             )
             citation_processor.update_citation_mapping(project_citation_mapping)
 
-        llm_step_result: LlmStepResult | None = None
+        llm_step_result = LlmStepResult(
+            reasoning=None,
+            answer=None,
+            tool_calls=None,
+            raw_answer=None,
+        )
 
         # Pass the total budget to construct_message_history, which will handle token allocation
         available_tokens = llm.config.max_input_tokens
@@ -1084,12 +1155,18 @@ def run_llm_loop(
                 # As long as 1 tool with citeable documents is called at any point, we ask the LLM to try to cite
                 should_cite_documents = True
 
-        if not llm_step_result or not llm_step_result.answer:
+        if not llm_step_result.answer and not llm_step_result.tool_calls:
+            raise _build_empty_llm_response_error(
+                llm=llm,
+                llm_step_result=llm_step_result,
+                tool_choice=tool_choice,
+            )
+
+        if not llm_step_result.answer:
             raise RuntimeError(
-                "The LLM did not return an answer. "
-                "Typically this is an issue with LLMs that do not support tool calling natively, "
-                "or the model serving API is not configured correctly. "
-                "This may also happen with models that are lower quality outputting invalid tool calls."
+                "The LLM did not return a final answer after tool execution. "
+                "Typically this indicates invalid tool-call output, a model/provider mismatch, "
+                "or serving API misconfiguration."
             )
 
         emitter.emit(

backend/onyx/chat/llm_step.py

@@ -1013,6 +1013,10 @@ def run_llm_step_pkt_generator(
     accumulated_reasoning = ""
     accumulated_answer = ""
     accumulated_raw_answer = ""
+    stream_chunk_count = 0
+    actionable_chunk_count = 0
+    empty_chunk_count = 0
+    finish_reasons: set[str] = set()
     xml_tool_call_content_filter = _XmlToolCallContentFilter()
 
     processor_state: Any = None
@@ -1145,6 +1149,7 @@ def run_llm_step_pkt_generator(
             user_identity=user_identity,
             timeout_override=timeout_override,
         ):
+            stream_chunk_count += 1
             if packet.usage:
                 usage = packet.usage
                 span_generation.span_data.usage = {
@@ -1154,16 +1159,21 @@ def run_llm_step_pkt_generator(
                     "cache_creation_input_tokens": usage.cache_creation_input_tokens,
                 }
                 # Note: LLM cost tracking is now handled in multi_llm.py
+            finish_reason = packet.choice.finish_reason
+            if finish_reason:
+                finish_reasons.add(str(finish_reason))
             delta = packet.choice.delta
 
             # Weird behavior from some model providers, just log and ignore for now
             if (
-                delta.content is None
+                not delta.content
                 and delta.reasoning_content is None
-                and delta.tool_calls is None
+                and not delta.tool_calls
             ):
+                empty_chunk_count += 1
                 logger.warning(
-                    f"LLM packet is empty (no contents, reasoning or tool calls). Skipping: {packet}"
+                    "LLM packet is empty (no content, reasoning, or tool calls). "
+                    f"finish_reason={finish_reason}. Skipping: {packet}"
                 )
                 continue
 
@@ -1172,6 +1182,8 @@ def run_llm_step_pkt_generator(
                     time.monotonic() - stream_start_time
                 )
                 first_action_recorded = True
+            if _delta_has_action(delta):
+                actionable_chunk_count += 1
 
             if custom_token_processor:
                 # The custom token processor can modify the deltas for specific custom logic
@@ -1307,6 +1319,15 @@ def run_llm_step_pkt_generator(
         else:
             logger.debug("Tool calls: []")
 
+    if actionable_chunk_count == 0:
+        logger.warning(
+            "LLM stream completed with no actionable deltas. "
+            f"chunks={stream_chunk_count}, empty_chunks={empty_chunk_count}, "
+            f"finish_reasons={sorted(finish_reasons)}, "
+            f"provider={llm.config.model_provider}, model={llm.config.model_name}, "
+            f"tool_choice={tool_choice}, tools_sent={len(tool_definitions)}"
+        )
+
     return (
         LlmStepResult(
             reasoning=accumulated_reasoning if accumulated_reasoning else None,

backend/onyx/chat/process_message.py

@@ -29,6 +29,7 @@ from onyx.chat.compression import compress_chat_history
 from onyx.chat.compression import find_summary_for_branch
 from onyx.chat.compression import get_compression_params
 from onyx.chat.emitter import get_default_emitter
+from onyx.chat.llm_loop import EmptyLLMResponseError
 from onyx.chat.llm_loop import run_llm_loop
 from onyx.chat.models import AnswerStream
 from onyx.chat.models import ChatBasicResponse
@@ -58,6 +59,7 @@ from onyx.db.chat import create_new_chat_message
 from onyx.db.chat import get_chat_session_by_id
 from onyx.db.chat import get_or_create_root_message
 from onyx.db.chat import reserve_message_id
+from onyx.db.enums import HookPoint
 from onyx.db.memory import get_memories
 from onyx.db.models import ChatMessage
 from onyx.db.models import ChatSession
@@ -67,11 +69,19 @@ from onyx.db.models import UserFile
 from onyx.db.projects import get_user_files_from_project
 from onyx.db.tools import get_tools
 from onyx.deep_research.dr_loop import run_deep_research_llm_loop
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import log_onyx_error
+from onyx.error_handling.exceptions import OnyxError
 from onyx.file_processing.extract_file_text import extract_file_text
 from onyx.file_store.models import ChatFileType
 from onyx.file_store.models import InMemoryChatFile
 from onyx.file_store.utils import load_in_memory_chat_files
 from onyx.file_store.utils import verify_user_files
+from onyx.hooks.executor import execute_hook
+from onyx.hooks.executor import HookSkipped
+from onyx.hooks.executor import HookSoftFailed
+from onyx.hooks.points.query_processing import QueryProcessingPayload
+from onyx.hooks.points.query_processing import QueryProcessingResponse
 from onyx.llm.factory import get_llm_for_persona
 from onyx.llm.factory import get_llm_token_counter
 from onyx.llm.interfaces import LLM
@@ -423,6 +433,32 @@ def determine_search_params(
     )
 
 
+def _resolve_query_processing_hook_result(
+    hook_result: BaseModel | HookSkipped | HookSoftFailed,
+    message_text: str,
+) -> str:
+    """Apply the Query Processing hook result to the message text.
+
+    Returns the (possibly rewritten) message text, or raises OnyxError with
+    QUERY_REJECTED if the hook signals rejection (query is null or empty).
+    HookSkipped and HookSoftFailed are pass-throughs — the original text is
+    returned unchanged.
+    """
+    if isinstance(hook_result, (HookSkipped, HookSoftFailed)):
+        return message_text
+    if not isinstance(hook_result, QueryProcessingResponse):
+        raise OnyxError(
+            OnyxErrorCode.INTERNAL_ERROR,
+            f"Expected QueryProcessingResponse from hook, got {type(hook_result).__name__}",
+        )
+    if not (hook_result.query and hook_result.query.strip()):
+        raise OnyxError(
+            OnyxErrorCode.QUERY_REJECTED,
+            hook_result.rejection_message or "Your query was rejected.",
+        )
+    return hook_result.query.strip()
+
+
 def handle_stream_message_objects(
     new_msg_req: SendMessageRequest,
     user: User,
@@ -483,6 +519,7 @@ def handle_stream_message_objects(
         persona = chat_session.persona
 
         message_text = new_msg_req.message
+
         user_identity = LLMUserIdentity(
             user_id=llm_user_identifier, session_id=str(chat_session.id)
         )
@@ -574,6 +611,27 @@ def handle_stream_message_objects(
         if parent_message.message_type == MessageType.USER:
             user_message = parent_message
         else:
+            # New message — run the Query Processing hook before saving to DB.
+            # Skipped on regeneration: the message already exists and was accepted previously.
+            # Skip the hook for empty/whitespace-only messages — no meaningful query
+            # to process, and SendMessageRequest.message has no min_length guard.
+            if message_text.strip():
+                hook_result = execute_hook(
+                    db_session=db_session,
+                    hook_point=HookPoint.QUERY_PROCESSING,
+                    payload=QueryProcessingPayload(
+                        query=message_text,
+                        # Pass None for anonymous users or authenticated users without an email
+                        # (e.g. some SSO flows). QueryProcessingPayload.user_email is str | None,
+                        # so None is accepted and serialised as null in both cases.
+                        user_email=None if user.is_anonymous else user.email,
+                        chat_session_id=str(chat_session.id),
+                    ).model_dump(),
+                )
+                message_text = _resolve_query_processing_hook_result(
+                    hook_result, message_text
+                )
+
             user_message = create_new_chat_message(
                 chat_session_id=chat_session.id,
                 parent_message=parent_message,
@@ -913,6 +971,17 @@ def handle_stream_message_objects(
                 state_container=state_container,
             )
 
+    except OnyxError as e:
+        if e.error_code is not OnyxErrorCode.QUERY_REJECTED:
+            log_onyx_error(e)
+        yield StreamingError(
+            error=e.detail,
+            error_code=e.error_code.code,
+            is_retryable=e.status_code >= 500,
+        )
+        db_session.rollback()
+        return
+
     except ValueError as e:
         logger.exception("Failed to process chat message.")
 
@@ -925,9 +994,28 @@ def handle_stream_message_objects(
         db_session.rollback()
         return
 
+    except EmptyLLMResponseError as e:
+        stack_trace = traceback.format_exc()
+
+        logger.warning(
+            "LLM returned an empty response "
+            f"(provider={e.provider}, model={e.model}, tool_choice={e.tool_choice})"
+        )
+
+        yield StreamingError(
+            error=e.client_error_msg,
+            stack_trace=stack_trace,
+            error_code=e.error_code,
+            is_retryable=e.is_retryable,
+            details={
+                "model": e.model,
+                "provider": e.provider,
+                "tool_choice": e.tool_choice.value,
+            },
+        )
+        db_session.rollback()
     except Exception as e:
         logger.exception(f"Failed to process chat message due to {e}")
-        error_msg = str(e)
         stack_trace = traceback.format_exc()
 
         if llm:
@@ -1046,10 +1134,46 @@ def llm_loop_completion_handle(
         )
 
 
-def remove_answer_citations(answer: str) -> str:
-    pattern = r"\s*\[\[\d+\]\]\(http[s]?://[^\s]+\)"
+_CITATION_LINK_START_PATTERN = re.compile(r"\s*\[\[\d+\]\]\(")
 
-    return re.sub(pattern, "", answer)
+
+def _find_markdown_link_end(text: str, destination_start: int) -> int | None:
+    depth = 0
+    i = destination_start
+
+    while i < len(text):
+        curr = text[i]
+        if curr == "\\":
+            i += 2
+            continue
+
+        if curr == "(":
+            depth += 1
+        elif curr == ")":
+            if depth == 0:
+                return i
+            depth -= 1
+
+        i += 1
+
+    return None
+
+
+def remove_answer_citations(answer: str) -> str:
+    stripped_parts: list[str] = []
+    cursor = 0
+
+    while match := _CITATION_LINK_START_PATTERN.search(answer, cursor):
+        stripped_parts.append(answer[cursor : match.start()])
+        link_end = _find_markdown_link_end(answer, match.end())
+        if link_end is None:
+            stripped_parts.append(answer[match.start() :])
+            return "".join(stripped_parts)
+
+        cursor = link_end + 1
+
+    stripped_parts.append(answer[cursor:])
+    return "".join(stripped_parts)
 
 
 @log_function_time()
@@ -1087,8 +1211,11 @@ def gather_stream(
         raise ValueError("Message ID is required")
 
     if answer is None:
-        # This should never be the case as these non-streamed flows do not have a stop-generation signal
-        raise RuntimeError("Answer was not generated")
+        if error_msg is not None:
+            answer = ""
+        else:
+            # This should never be the case as these non-streamed flows do not have a stop-generation signal
+            raise RuntimeError("Answer was not generated")
 
     return ChatBasicResponse(
         answer=answer,

backend/onyx/configs/app_configs.py

@@ -278,14 +278,17 @@ USING_AWS_MANAGED_OPENSEARCH = (
 OPENSEARCH_PROFILING_DISABLED = (
     os.environ.get("OPENSEARCH_PROFILING_DISABLED", "").lower() == "true"
 )
-
+# Whether to disable match highlights for OpenSearch. Defaults to True for now
+# as we investigate query performance.
+OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED = (
+    os.environ.get("OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED", "true").lower() == "true"
+)
 # When enabled, OpenSearch returns detailed score breakdowns for each hit.
 # Useful for debugging and tuning search relevance. Has ~10-30% performance overhead according to documentation.
 # Seems for Hybrid Search in practice, the impact is actually more like 1000x slower.
 OPENSEARCH_EXPLAIN_ENABLED = (
     os.environ.get("OPENSEARCH_EXPLAIN_ENABLED", "").lower() == "true"
 )
-
 # Analyzer used for full-text fields (title, content). Use OpenSearch built-in analyzer
 # names (e.g. "english", "standard", "german"). Affects stemming and tokenization;
 # existing indices need reindexing after a change.
@@ -318,8 +321,16 @@ VERIFY_CREATE_OPENSEARCH_INDEX_ON_INIT_MT = (
 OPENSEARCH_MIGRATION_GET_VESPA_CHUNKS_PAGE_SIZE = int(
     os.environ.get("OPENSEARCH_MIGRATION_GET_VESPA_CHUNKS_PAGE_SIZE") or 500
 )
-OPENSEARCH_OVERRIDE_DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES = int(
-    os.environ.get("OPENSEARCH_DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES") or 0
+# If set, will override the default number of shards and replicas for the index.
+OPENSEARCH_INDEX_NUM_SHARDS: int | None = (
+    int(os.environ["OPENSEARCH_INDEX_NUM_SHARDS"])
+    if os.environ.get("OPENSEARCH_INDEX_NUM_SHARDS", None) is not None
+    else None
+)
+OPENSEARCH_INDEX_NUM_REPLICAS: int | None = (
+    int(os.environ["OPENSEARCH_INDEX_NUM_REPLICAS"])
+    if os.environ.get("OPENSEARCH_INDEX_NUM_REPLICAS", None) is not None
+    else None
 )
 
 VESPA_HOST = os.environ.get("VESPA_HOST") or "localhost"

backend/onyx/configs/constants.py

@@ -177,6 +177,14 @@ USER_FILE_PROJECT_SYNC_MAX_QUEUE_DEPTH = 500
 
 CELERY_USER_FILE_PROJECT_SYNC_LOCK_TIMEOUT = 5 * 60  # 5 minutes (in seconds)
 
+# How long a queued user-file-delete task is valid before workers discard it.
+# Mirrors the processing task expiry to prevent indefinite queue growth when
+# files are stuck in DELETING status and the beat keeps re-enqueuing them.
+CELERY_USER_FILE_DELETE_TASK_EXPIRES = 60  # 1 minute (in seconds)
+
+# Max queue depth before the delete beat stops enqueuing more delete tasks.
+USER_FILE_DELETE_MAX_QUEUE_DEPTH = 500
+
 CELERY_SANDBOX_FILE_SYNC_LOCK_TIMEOUT = 5 * 60  # 5 minutes (in seconds)
 
 DANSWER_REDIS_FUNCTION_LOCK_PREFIX = "da_function_lock:"
@@ -469,6 +477,9 @@ class OnyxRedisLocks:
     USER_FILE_PROJECT_SYNC_QUEUED_PREFIX = "da_lock:user_file_project_sync_queued"
     USER_FILE_DELETE_BEAT_LOCK = "da_lock:check_user_file_delete_beat"
     USER_FILE_DELETE_LOCK_PREFIX = "da_lock:user_file_delete"
+    # Short-lived key set when a delete task is enqueued; cleared when the worker picks it up.
+    # Prevents the beat from re-enqueuing the same file while a delete task is already queued.
+    USER_FILE_DELETE_QUEUED_PREFIX = "da_lock:user_file_delete_queued"
 
     # Release notes
     RELEASE_NOTES_FETCH_LOCK = "da_lock:release_notes_fetch"
@@ -597,6 +608,9 @@ class OnyxCeleryTask:
     EXPORT_QUERY_HISTORY_TASK = "export_query_history_task"
     EXPORT_QUERY_HISTORY_CLEANUP_TASK = "export_query_history_cleanup_task"
 
+    # Hook execution log retention
+    HOOK_EXECUTION_LOG_CLEANUP_TASK = "hook_execution_log_cleanup_task"
+
     # Sandbox cleanup
     CLEANUP_IDLE_SANDBOXES = "cleanup_idle_sandboxes"
     CLEANUP_OLD_SNAPSHOTS = "cleanup_old_snapshots"

backend/onyx/connectors/google_utils/google_utils.py

@@ -157,9 +157,7 @@ def _execute_single_retrieval(
             logger.error(f"Error executing request: {e}")
             raise e
         elif _is_rate_limit_error(e):
-            results = _execute_with_retry(
-                lambda: retrieval_function(**request_kwargs).execute()
-            )
+            results = _execute_with_retry(retrieval_function(**request_kwargs))
         elif e.resp.status == 404 or e.resp.status == 403:
             if continue_on_404_or_403:
                 logger.debug(f"Error executing request: {e}")

backend/onyx/db/connector_credential_pair.py

@@ -511,7 +511,7 @@ def add_credential_to_connector(
     user: User,
     connector_id: int,
     credential_id: int,
-    cc_pair_name: str | None,
+    cc_pair_name: str,
     access_type: AccessType,
     groups: list[int] | None,
     auto_sync_options: dict | None = None,

backend/onyx/db/hook.py

@@ -0,0 +1,233 @@
+import datetime
+from uuid import UUID
+
+from sqlalchemy import delete
+from sqlalchemy import select
+from sqlalchemy.engine import CursorResult
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.orm import selectinload
+from sqlalchemy.orm import Session
+
+from onyx.db.constants import UNSET
+from onyx.db.constants import UnsetType
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.db.models import Hook
+from onyx.db.models import HookExecutionLog
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
+
+
+# ── Hook CRUD ────────────────────────────────────────────────────────────
+
+
+def get_hook_by_id(
+    *,
+    db_session: Session,
+    hook_id: int,
+    include_deleted: bool = False,
+    include_creator: bool = False,
+) -> Hook | None:
+    stmt = select(Hook).where(Hook.id == hook_id)
+    if not include_deleted:
+        stmt = stmt.where(Hook.deleted.is_(False))
+    if include_creator:
+        stmt = stmt.options(selectinload(Hook.creator))
+    return db_session.scalar(stmt)
+
+
+def get_non_deleted_hook_by_hook_point(
+    *,
+    db_session: Session,
+    hook_point: HookPoint,
+    include_creator: bool = False,
+) -> Hook | None:
+    stmt = (
+        select(Hook).where(Hook.hook_point == hook_point).where(Hook.deleted.is_(False))
+    )
+    if include_creator:
+        stmt = stmt.options(selectinload(Hook.creator))
+    return db_session.scalar(stmt)
+
+
+def get_hooks(
+    *,
+    db_session: Session,
+    include_deleted: bool = False,
+    include_creator: bool = False,
+) -> list[Hook]:
+    stmt = select(Hook)
+    if not include_deleted:
+        stmt = stmt.where(Hook.deleted.is_(False))
+    if include_creator:
+        stmt = stmt.options(selectinload(Hook.creator))
+    stmt = stmt.order_by(Hook.hook_point, Hook.created_at.desc())
+    return list(db_session.scalars(stmt).all())
+
+
+def create_hook__no_commit(
+    *,
+    db_session: Session,
+    name: str,
+    hook_point: HookPoint,
+    endpoint_url: str | None = None,
+    api_key: str | None = None,
+    fail_strategy: HookFailStrategy,
+    timeout_seconds: float,
+    is_active: bool = False,
+    creator_id: UUID | None = None,
+) -> Hook:
+    """Create a new hook for the given hook point.
+
+    At most one non-deleted hook per hook point is allowed. Raises
+    OnyxError(CONFLICT) if a hook already exists, including under concurrent
+    duplicate creates where the partial unique index fires an IntegrityError.
+    """
+    existing = get_non_deleted_hook_by_hook_point(
+        db_session=db_session, hook_point=hook_point
+    )
+    if existing:
+        raise OnyxError(
+            OnyxErrorCode.CONFLICT,
+            f"A hook for '{hook_point.value}' already exists (id={existing.id}).",
+        )
+
+    hook = Hook(
+        name=name,
+        hook_point=hook_point,
+        endpoint_url=endpoint_url,
+        api_key=api_key,
+        fail_strategy=fail_strategy,
+        timeout_seconds=timeout_seconds,
+        is_active=is_active,
+        creator_id=creator_id,
+    )
+    # Use a savepoint so that a failed insert only rolls back this operation,
+    # not the entire outer transaction.
+    savepoint = db_session.begin_nested()
+    try:
+        db_session.add(hook)
+        savepoint.commit()
+    except IntegrityError as exc:
+        savepoint.rollback()
+        if "ix_hook_one_non_deleted_per_point" in str(exc.orig):
+            raise OnyxError(
+                OnyxErrorCode.CONFLICT,
+                f"A hook for '{hook_point.value}' already exists.",
+            )
+        raise  # re-raise unrelated integrity errors (FK violations, etc.)
+    return hook
+
+
+def update_hook__no_commit(
+    *,
+    db_session: Session,
+    hook_id: int,
+    name: str | None = None,
+    endpoint_url: str | None | UnsetType = UNSET,
+    api_key: str | None | UnsetType = UNSET,
+    fail_strategy: HookFailStrategy | None = None,
+    timeout_seconds: float | None = None,
+    is_active: bool | None = None,
+    is_reachable: bool | None = None,
+    include_creator: bool = False,
+) -> Hook:
+    """Update hook fields.
+
+    Sentinel conventions:
+    - endpoint_url, api_key: pass UNSET to leave unchanged; pass None to clear.
+    - name, fail_strategy, timeout_seconds, is_active, is_reachable: pass None to leave unchanged.
+    """
+    hook = get_hook_by_id(
+        db_session=db_session, hook_id=hook_id, include_creator=include_creator
+    )
+    if hook is None:
+        raise OnyxError(OnyxErrorCode.NOT_FOUND, f"Hook with id {hook_id} not found.")
+
+    if name is not None:
+        hook.name = name
+    if not isinstance(endpoint_url, UnsetType):
+        hook.endpoint_url = endpoint_url
+    if not isinstance(api_key, UnsetType):
+        hook.api_key = api_key  # type: ignore[assignment]  # EncryptedString coerces str → SensitiveValue at the ORM level
+    if fail_strategy is not None:
+        hook.fail_strategy = fail_strategy
+    if timeout_seconds is not None:
+        hook.timeout_seconds = timeout_seconds
+    if is_active is not None:
+        hook.is_active = is_active
+    if is_reachable is not None:
+        hook.is_reachable = is_reachable
+
+    db_session.flush()
+    return hook
+
+
+def delete_hook__no_commit(
+    *,
+    db_session: Session,
+    hook_id: int,
+) -> None:
+    hook = get_hook_by_id(db_session=db_session, hook_id=hook_id)
+    if hook is None:
+        raise OnyxError(OnyxErrorCode.NOT_FOUND, f"Hook with id {hook_id} not found.")
+
+    hook.deleted = True
+    hook.is_active = False
+    db_session.flush()
+
+
+# ── HookExecutionLog CRUD ────────────────────────────────────────────────
+
+
+def create_hook_execution_log__no_commit(
+    *,
+    db_session: Session,
+    hook_id: int,
+    is_success: bool,
+    error_message: str | None = None,
+    status_code: int | None = None,
+    duration_ms: int | None = None,
+) -> HookExecutionLog:
+    log = HookExecutionLog(
+        hook_id=hook_id,
+        is_success=is_success,
+        error_message=error_message,
+        status_code=status_code,
+        duration_ms=duration_ms,
+    )
+    db_session.add(log)
+    db_session.flush()
+    return log
+
+
+def get_hook_execution_logs(
+    *,
+    db_session: Session,
+    hook_id: int,
+    limit: int,
+) -> list[HookExecutionLog]:
+    stmt = (
+        select(HookExecutionLog)
+        .where(HookExecutionLog.hook_id == hook_id)
+        .order_by(HookExecutionLog.created_at.desc())
+        .limit(limit)
+    )
+    return list(db_session.scalars(stmt).all())
+
+
+def cleanup_old_execution_logs__no_commit(
+    *,
+    db_session: Session,
+    max_age_days: int,
+) -> int:
+    """Delete execution logs older than max_age_days. Returns the number of rows deleted."""
+    cutoff = datetime.datetime.now(datetime.timezone.utc) - datetime.timedelta(
+        days=max_age_days
+    )
+    result: CursorResult = db_session.execute(  # type: ignore[assignment]
+        delete(HookExecutionLog)
+        .where(HookExecutionLog.created_at < cutoff)
+        .execution_options(synchronize_session=False)
+    )
+    return result.rowcount

backend/onyx/db/projects.py

@@ -12,6 +12,7 @@ from sqlalchemy.orm import Session
 from starlette.background import BackgroundTasks
 
 from onyx.configs.app_configs import DISABLE_VECTOR_DB
+from onyx.configs.constants import CELERY_USER_FILE_PROCESSING_TASK_EXPIRES
 from onyx.configs.constants import FileOrigin
 from onyx.configs.constants import OnyxCeleryPriority
 from onyx.configs.constants import OnyxCeleryQueues
@@ -144,6 +145,7 @@ def upload_files_to_user_files_with_indexing(
                 kwargs={"user_file_id": user_file.id, "tenant_id": tenant_id},
                 queue=OnyxCeleryQueues.USER_FILE_PROCESSING,
                 priority=OnyxCeleryPriority.HIGH,
+                expires=CELERY_USER_FILE_PROCESSING_TASK_EXPIRES,
             )
             logger.info(
                 f"Triggered indexing for user_file_id={user_file.id} with task_id={task.id}"

backend/onyx/db/swap_index.py

@@ -2,6 +2,7 @@ import time
 
 from sqlalchemy.orm import Session
 
+from onyx.configs.app_configs import DISABLE_VECTOR_DB
 from onyx.configs.app_configs import VESPA_NUM_ATTEMPTS_ON_STARTUP
 from onyx.configs.constants import KV_REINDEX_KEY
 from onyx.db.connector_credential_pair import get_connector_credential_pairs
@@ -149,6 +150,9 @@ def check_and_perform_index_swap(db_session: Session) -> SearchSettings | None:
     Returns None if search settings did not change, or the old search settings if they
     did change.
     """
+    if DISABLE_VECTOR_DB:
+        return None
+
     # Default CC-pair created for Ingestion API unused here
     all_cc_pairs = get_connector_credential_pairs(db_session)
     cc_pair_count = max(len(all_cc_pairs) - 1, 0)

backend/onyx/document_index/opensearch/client.py

@@ -1,3 +1,4 @@
+import json
 import logging
 import time
 from contextlib import AbstractContextManager
@@ -18,6 +19,7 @@ from onyx.configs.app_configs import OPENSEARCH_HOST
 from onyx.configs.app_configs import OPENSEARCH_REST_API_PORT
 from onyx.document_index.interfaces_new import TenantState
 from onyx.document_index.opensearch.schema import DocumentChunk
+from onyx.document_index.opensearch.schema import DocumentChunkWithoutVectors
 from onyx.document_index.opensearch.schema import get_opensearch_doc_chunk_id
 from onyx.document_index.opensearch.search import DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW
 from onyx.utils.logger import setup_logger
@@ -56,8 +58,8 @@ class SearchHit(BaseModel, Generic[SchemaDocumentModel]):
     # Maps schema property name to a list of highlighted snippets with match
     # terms wrapped in tags (e.g. "something <hi>keyword</hi> other thing").
     match_highlights: dict[str, list[str]] = {}
-    # Score explanation from OpenSearch when "explain": true is set in the query.
-    # Contains detailed breakdown of how the score was calculated.
+    # Score explanation from OpenSearch when "explain": true is set in the
+    # query. Contains detailed breakdown of how the score was calculated.
     explanation: dict[str, Any] | None = None
 
 
@@ -833,9 +835,13 @@ class OpenSearchIndexClient(OpenSearchClient):
     @log_function_time(print_only=True, debug_only=True)
     def search(
         self, body: dict[str, Any], search_pipeline_id: str | None
-    ) -> list[SearchHit[DocumentChunk]]:
+    ) -> list[SearchHit[DocumentChunkWithoutVectors]]:
         """Searches the index.
 
+        NOTE: Does not return vector fields. In order to take advantage of
+        performance benefits, the search body should exclude the schema's vector
+        fields.
+
         TODO(andrei): Ideally we could check that every field in the body is
         present in the index, to avoid a class of runtime bugs that could easily
         be caught during development. Or change the function signature to accept
@@ -883,7 +889,7 @@ class OpenSearchIndexClient(OpenSearchClient):
             raise_on_timeout=True,
         )
 
-        search_hits: list[SearchHit[DocumentChunk]] = []
+        search_hits: list[SearchHit[DocumentChunkWithoutVectors]] = []
         for hit in hits:
             document_chunk_source: dict[str, Any] | None = hit.get("_source")
             if not document_chunk_source:
@@ -893,8 +899,10 @@ class OpenSearchIndexClient(OpenSearchClient):
             document_chunk_score = hit.get("_score", None)
             match_highlights: dict[str, list[str]] = hit.get("highlight", {})
             explanation: dict[str, Any] | None = hit.get("_explanation", None)
-            search_hit = SearchHit[DocumentChunk](
-                document_chunk=DocumentChunk.model_validate(document_chunk_source),
+            search_hit = SearchHit[DocumentChunkWithoutVectors](
+                document_chunk=DocumentChunkWithoutVectors.model_validate(
+                    document_chunk_source
+                ),
                 score=document_chunk_score,
                 match_highlights=match_highlights,
                 explanation=explanation,
@@ -1055,7 +1063,7 @@ class OpenSearchIndexClient(OpenSearchClient):
                 f"Body: {get_new_body_without_vectors(body)}\n"
                 f"Search pipeline ID: {search_pipeline_id}\n"
                 f"Phase took: {phase_took}\n"
-                f"Profile: {profile}\n"
+                f"Profile: {json.dumps(profile, indent=2)}\n"
             )
         if timed_out:
             error_str = f"OpenSearch client error: Search timed out for index {self._index_name}."

backend/onyx/document_index/opensearch/constants.py

@@ -1,12 +1,23 @@
 # Default value for the maximum number of tokens a chunk can hold, if none is
 # specified when creating an index.
-from onyx.configs.app_configs import (
-    OPENSEARCH_OVERRIDE_DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES,
-)
+import os
+from enum import Enum
 
 
 DEFAULT_MAX_CHUNK_SIZE = 512
 
+
+# By default OpenSearch will only return a maximum of this many results in a
+# given search. This value is configurable in the index settings.
+DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW = 10_000
+
+
+# For documents which do not have a value for LAST_UPDATED_FIELD_NAME, we assume
+# that the document was last updated this many days ago for the purpose of time
+# cutoff filtering during retrieval.
+ASSUMED_DOCUMENT_AGE_DAYS = 90
+
+
 # Size of the dynamic list used to consider elements during kNN graph creation.
 # Higher values improve search quality but increase indexing time. Values
 # typically range between 100 - 512.
@@ -26,10 +37,10 @@ M = 32  # Set relatively high for better accuracy.
 # we have a much higher chance of all 10 of the final desired docs showing up
 # and getting scored. In worse situations, the final 10 docs don't even show up
 # as the final 10 (worse than just a miss at the reranking step).
-DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES = (
-    OPENSEARCH_OVERRIDE_DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES
-    if OPENSEARCH_OVERRIDE_DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES > 0
-    else 750
+# Defaults to 100 for now. Initially this defaulted to 750 but we were seeing
+# poor search performance.
+DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES = int(
+    os.environ.get("DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES", 100)
 )
 
 # Number of vectors to examine to decide the top k neighbors for the HNSW
@@ -39,23 +50,43 @@ DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES = (
 # larger than k, you can provide the size parameter to limit the final number of
 # results to k." from
 # https://docs.opensearch.org/latest/query-dsl/specialized/k-nn/index/#ef_search
-EF_SEARCH = DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES
+EF_SEARCH = DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES
 
-# Since the titles are included in the contents, the embedding matches are
-# heavily downweighted as they act as a boost rather than an independent scoring
-# component.
-SEARCH_TITLE_VECTOR_WEIGHT = 0.1
-SEARCH_CONTENT_VECTOR_WEIGHT = 0.45
-# Single keyword weight for both title and content (merged from former title
-# keyword + content keyword).
-SEARCH_KEYWORD_WEIGHT = 0.45
 
-# NOTE: It is critical that the order of these weights matches the order of the
-# sub-queries in the hybrid search.
-HYBRID_SEARCH_NORMALIZATION_WEIGHTS = [
-    SEARCH_TITLE_VECTOR_WEIGHT,
-    SEARCH_CONTENT_VECTOR_WEIGHT,
-    SEARCH_KEYWORD_WEIGHT,
-]
+class HybridSearchSubqueryConfiguration(Enum):
+    TITLE_VECTOR_CONTENT_VECTOR_TITLE_CONTENT_COMBINED_KEYWORD = 1
+    # Current default.
+    CONTENT_VECTOR_TITLE_CONTENT_COMBINED_KEYWORD = 2
 
-assert sum(HYBRID_SEARCH_NORMALIZATION_WEIGHTS) == 1.0
+
+# Will raise and block application start if HYBRID_SEARCH_SUBQUERY_CONFIGURATION
+# is set but not a valid value. If not set, defaults to
+# CONTENT_VECTOR_TITLE_CONTENT_COMBINED_KEYWORD.
+HYBRID_SEARCH_SUBQUERY_CONFIGURATION: HybridSearchSubqueryConfiguration = (
+    HybridSearchSubqueryConfiguration(
+        int(os.environ["HYBRID_SEARCH_SUBQUERY_CONFIGURATION"])
+    )
+    if os.environ.get("HYBRID_SEARCH_SUBQUERY_CONFIGURATION", None) is not None
+    else HybridSearchSubqueryConfiguration.CONTENT_VECTOR_TITLE_CONTENT_COMBINED_KEYWORD
+)
+
+
+class HybridSearchNormalizationPipeline(Enum):
+    # Current default.
+    MIN_MAX = 1
+    # NOTE: Using z-score normalization is better for hybrid search from a
+    # theoretical standpoint. Empirically on a small dataset of up to 10K docs,
+    # it's not very different. Likely more impactful at scale.
+    # https://opensearch.org/blog/introducing-the-z-score-normalization-technique-for-hybrid-search/
+    ZSCORE = 2
+
+
+# Will raise and block application start if HYBRID_SEARCH_NORMALIZATION_PIPELINE
+# is set but not a valid value. If not set, defaults to MIN_MAX.
+HYBRID_SEARCH_NORMALIZATION_PIPELINE: HybridSearchNormalizationPipeline = (
+    HybridSearchNormalizationPipeline(
+        int(os.environ["HYBRID_SEARCH_NORMALIZATION_PIPELINE"])
+    )
+    if os.environ.get("HYBRID_SEARCH_NORMALIZATION_PIPELINE", None) is not None
+    else HybridSearchNormalizationPipeline.MIN_MAX
+)

backend/onyx/document_index/opensearch/opensearch_document_index.py

@@ -47,6 +47,7 @@ from onyx.document_index.opensearch.schema import ACCESS_CONTROL_LIST_FIELD_NAME
 from onyx.document_index.opensearch.schema import CONTENT_FIELD_NAME
 from onyx.document_index.opensearch.schema import DOCUMENT_SETS_FIELD_NAME
 from onyx.document_index.opensearch.schema import DocumentChunk
+from onyx.document_index.opensearch.schema import DocumentChunkWithoutVectors
 from onyx.document_index.opensearch.schema import DocumentSchema
 from onyx.document_index.opensearch.schema import get_opensearch_doc_chunk_id
 from onyx.document_index.opensearch.schema import GLOBAL_BOOST_FIELD_NAME
@@ -55,16 +56,13 @@ from onyx.document_index.opensearch.schema import PERSONAS_FIELD_NAME
 from onyx.document_index.opensearch.schema import USER_PROJECTS_FIELD_NAME
 from onyx.document_index.opensearch.search import DocumentQuery
 from onyx.document_index.opensearch.search import (
-    MIN_MAX_NORMALIZATION_PIPELINE_CONFIG,
+    get_min_max_normalization_pipeline_name_and_config,
 )
 from onyx.document_index.opensearch.search import (
-    MIN_MAX_NORMALIZATION_PIPELINE_NAME,
+    get_normalization_pipeline_name_and_config,
 )
 from onyx.document_index.opensearch.search import (
-    ZSCORE_NORMALIZATION_PIPELINE_CONFIG,
-)
-from onyx.document_index.opensearch.search import (
-    ZSCORE_NORMALIZATION_PIPELINE_NAME,
+    get_zscore_normalization_pipeline_name_and_config,
 )
 from onyx.indexing.models import DocMetadataAwareIndexChunk
 from onyx.indexing.models import Document
@@ -103,18 +101,24 @@ def set_cluster_state(client: OpenSearchClient) -> None:
             "is not the first time running Onyx against this instance of OpenSearch, these "
             "settings have likely already been set. Not taking any further action..."
         )
-    client.create_search_pipeline(
-        pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME,
-        pipeline_body=MIN_MAX_NORMALIZATION_PIPELINE_CONFIG,
+    min_max_normalization_pipeline_name, min_max_normalization_pipeline_config = (
+        get_min_max_normalization_pipeline_name_and_config()
+    )
+    zscore_normalization_pipeline_name, zscore_normalization_pipeline_config = (
+        get_zscore_normalization_pipeline_name_and_config()
     )
     client.create_search_pipeline(
-        pipeline_id=ZSCORE_NORMALIZATION_PIPELINE_NAME,
-        pipeline_body=ZSCORE_NORMALIZATION_PIPELINE_CONFIG,
+        pipeline_id=min_max_normalization_pipeline_name,
+        pipeline_body=min_max_normalization_pipeline_config,
+    )
+    client.create_search_pipeline(
+        pipeline_id=zscore_normalization_pipeline_name,
+        pipeline_body=zscore_normalization_pipeline_config,
     )
 
 
 def _convert_retrieved_opensearch_chunk_to_inference_chunk_uncleaned(
-    chunk: DocumentChunk,
+    chunk: DocumentChunkWithoutVectors,
     score: float | None,
     highlights: dict[str, list[str]],
 ) -> InferenceChunkUncleaned:
@@ -877,7 +881,7 @@ class OpenSearchDocumentIndex(DocumentIndex):
         )
         results: list[InferenceChunk] = []
         for chunk_request in chunk_requests:
-            search_hits: list[SearchHit[DocumentChunk]] = []
+            search_hits: list[SearchHit[DocumentChunkWithoutVectors]] = []
             query_body = DocumentQuery.get_from_document_id_query(
                 document_id=chunk_request.document_id,
                 tenant_state=self._tenant_state,
@@ -940,17 +944,92 @@ class OpenSearchDocumentIndex(DocumentIndex):
             index_filters=filters,
             include_hidden=False,
         )
-        # NOTE: Using z-score normalization here because it's better for hybrid
-        # search from a theoretical standpoint. Empirically on a small dataset
-        # of up to 10K docs, it's not very different. Likely more impactful at
-        # scale.
-        # https://opensearch.org/blog/introducing-the-z-score-normalization-technique-for-hybrid-search/
-        search_hits: list[SearchHit[DocumentChunk]] = self._client.search(
+        normalization_pipeline_name, _ = get_normalization_pipeline_name_and_config()
+        search_hits: list[SearchHit[DocumentChunkWithoutVectors]] = self._client.search(
             body=query_body,
-            search_pipeline_id=ZSCORE_NORMALIZATION_PIPELINE_NAME,
+            search_pipeline_id=normalization_pipeline_name,
+        )
+
+        # Good place for a breakpoint to inspect the search hits if you have
+        # "explain" enabled.
+        inference_chunks_uncleaned: list[InferenceChunkUncleaned] = [
+            _convert_retrieved_opensearch_chunk_to_inference_chunk_uncleaned(
+                search_hit.document_chunk, search_hit.score, search_hit.match_highlights
+            )
+            for search_hit in search_hits
+        ]
+        inference_chunks: list[InferenceChunk] = cleanup_content_for_chunks(
+            inference_chunks_uncleaned
+        )
+
+        return inference_chunks
+
+    def keyword_retrieval(
+        self,
+        query: str,
+        filters: IndexFilters,
+        num_to_retrieve: int,
+    ) -> list[InferenceChunk]:
+        logger.debug(
+            f"[OpenSearchDocumentIndex] Keyword retrieving {num_to_retrieve} chunks for index {self._index_name}."
+        )
+        query_body = DocumentQuery.get_keyword_search_query(
+            query_text=query,
+            num_hits=num_to_retrieve,
+            tenant_state=self._tenant_state,
+            # NOTE: Index filters includes metadata tags which were filtered
+            # for invalid unicode at indexing time. In theory it would be
+            # ideal to do filtering here as well, in practice we never did
+            # that in the Vespa codepath and have not seen issues in
+            # production, so we deliberately conform to the existing logic
+            # in order to not unknowningly introduce a possible bug.
+            index_filters=filters,
+            include_hidden=False,
+        )
+        search_hits: list[SearchHit[DocumentChunkWithoutVectors]] = self._client.search(
+            body=query_body,
+            search_pipeline_id=None,
+        )
+
+        inference_chunks_uncleaned: list[InferenceChunkUncleaned] = [
+            _convert_retrieved_opensearch_chunk_to_inference_chunk_uncleaned(
+                search_hit.document_chunk, search_hit.score, search_hit.match_highlights
+            )
+            for search_hit in search_hits
+        ]
+        inference_chunks: list[InferenceChunk] = cleanup_content_for_chunks(
+            inference_chunks_uncleaned
+        )
+
+        return inference_chunks
+
+    def semantic_retrieval(
+        self,
+        query_embedding: Embedding,
+        filters: IndexFilters,
+        num_to_retrieve: int,
+    ) -> list[InferenceChunk]:
+        logger.debug(
+            f"[OpenSearchDocumentIndex] Semantic retrieving {num_to_retrieve} chunks for index {self._index_name}."
+        )
+        query_body = DocumentQuery.get_semantic_search_query(
+            query_embedding=query_embedding,
+            num_hits=num_to_retrieve,
+            tenant_state=self._tenant_state,
+            # NOTE: Index filters includes metadata tags which were filtered
+            # for invalid unicode at indexing time. In theory it would be
+            # ideal to do filtering here as well, in practice we never did
+            # that in the Vespa codepath and have not seen issues in
+            # production, so we deliberately conform to the existing logic
+            # in order to not unknowningly introduce a possible bug.
+            index_filters=filters,
+            include_hidden=False,
+        )
+        search_hits: list[SearchHit[DocumentChunkWithoutVectors]] = self._client.search(
+            body=query_body,
+            search_pipeline_id=None,
         )
 
-        # Good place for a breakpoint to inspect the search hits if you have "explain" enabled.
         inference_chunks_uncleaned: list[InferenceChunkUncleaned] = [
             _convert_retrieved_opensearch_chunk_to_inference_chunk_uncleaned(
                 search_hit.document_chunk, search_hit.score, search_hit.match_highlights
@@ -977,7 +1056,7 @@ class OpenSearchDocumentIndex(DocumentIndex):
             index_filters=filters,
             num_to_retrieve=num_to_retrieve,
         )
-        search_hits: list[SearchHit[DocumentChunk]] = self._client.search(
+        search_hits: list[SearchHit[DocumentChunkWithoutVectors]] = self._client.search(
             body=query_body,
             search_pipeline_id=None,
         )

backend/onyx/document_index/opensearch/schema.py

@@ -11,6 +11,8 @@ from pydantic import model_serializer
 from pydantic import model_validator
 from pydantic import SerializerFunctionWrapHandler
 
+from onyx.configs.app_configs import OPENSEARCH_INDEX_NUM_REPLICAS
+from onyx.configs.app_configs import OPENSEARCH_INDEX_NUM_SHARDS
 from onyx.configs.app_configs import OPENSEARCH_TEXT_ANALYZER
 from onyx.configs.app_configs import USING_AWS_MANAGED_OPENSEARCH
 from onyx.document_index.interfaces_new import TenantState
@@ -100,9 +102,9 @@ def set_or_convert_timezone_to_utc(value: datetime) -> datetime:
     return value
 
 
-class DocumentChunk(BaseModel):
+class DocumentChunkWithoutVectors(BaseModel):
     """
-    Represents a chunk of a document in the OpenSearch index.
+    Represents a chunk of a document in the OpenSearch index without vectors.
 
     The names of these fields are based on the OpenSearch schema. Changes to the
     schema require changes here. See get_document_schema.
@@ -124,9 +126,7 @@ class DocumentChunk(BaseModel):
 
     # Either both should be None or both should be non-None.
     title: str | None = None
-    title_vector: list[float] | None = None
     content: str
-    content_vector: list[float]
 
     source_type: str
     # A list of key-value pairs separated by INDEX_SEPARATOR. See
@@ -176,19 +176,9 @@ class DocumentChunk(BaseModel):
     def __str__(self) -> str:
         return (
             f"DocumentChunk(document_id={self.document_id}, chunk_index={self.chunk_index}, "
-            f"content length={len(self.content)}, content vector length={len(self.content_vector)}, "
-            f"tenant_id={self.tenant_id.tenant_id})"
+            f"content length={len(self.content)}, tenant_id={self.tenant_id.tenant_id})."
         )
 
-    @model_validator(mode="after")
-    def check_title_and_title_vector_are_consistent(self) -> Self:
-        # title and title_vector should both either be None or not.
-        if self.title is not None and self.title_vector is None:
-            raise ValueError("Bug: Title vector must not be None if title is not None.")
-        if self.title_vector is not None and self.title is None:
-            raise ValueError("Bug: Title must not be None if title vector is not None.")
-        return self
-
     @model_serializer(mode="wrap")
     def serialize_model(
         self, handler: SerializerFunctionWrapHandler
@@ -305,6 +295,35 @@ class DocumentChunk(BaseModel):
             return TenantState(tenant_id=value, multitenant=MULTI_TENANT)
 
 
+class DocumentChunk(DocumentChunkWithoutVectors):
+    """Represents a chunk of a document in the OpenSearch index.
+
+    The names of these fields are based on the OpenSearch schema. Changes to the
+    schema require changes here. See get_document_schema.
+    """
+
+    model_config = {"frozen": True}
+
+    title_vector: list[float] | None = None
+    content_vector: list[float]
+
+    def __str__(self) -> str:
+        return (
+            f"DocumentChunk(document_id={self.document_id}, chunk_index={self.chunk_index}, "
+            f"content length={len(self.content)}, content vector length={len(self.content_vector)}, "
+            f"tenant_id={self.tenant_id.tenant_id})"
+        )
+
+    @model_validator(mode="after")
+    def check_title_and_title_vector_are_consistent(self) -> Self:
+        # title and title_vector should both either be None or not.
+        if self.title is not None and self.title_vector is None:
+            raise ValueError("Bug: Title vector must not be None if title is not None.")
+        if self.title_vector is not None and self.title is None:
+            raise ValueError("Bug: Title must not be None if title vector is not None.")
+        return self
+
+
 class DocumentSchema:
     """
     Represents the schema and indexing strategies of the OpenSearch index.
@@ -516,78 +535,35 @@ class DocumentSchema:
 
         return schema
 
-    @staticmethod
-    def get_index_settings() -> dict[str, Any]:
-        """
-        Standard settings for reasonable local index and search performance.
-        """
-        return {
-            "index": {
-                "number_of_shards": 1,
-                "number_of_replicas": 1,
-                # Required for vector search.
-                "knn": True,
-                "knn.algo_param.ef_search": EF_SEARCH,
-            }
-        }
-
-    @staticmethod
-    def get_index_settings_for_aws_managed_opensearch_st_dev() -> dict[str, Any]:
-        """
-        Settings for AWS-managed OpenSearch.
-
-        Our AWS-managed OpenSearch cluster has 3 data nodes in 3 availability
-        zones.
-          - We use 3 shards to distribute load across all data nodes.
-          - We use 2 replicas to ensure each shard has a copy in each
-            availability zone. This is a hard requirement from AWS. The number
-            of data copies, including the primary (not a replica) copy, must be
-            divisible by the number of AZs.
-        """
-        return {
-            "index": {
-                "number_of_shards": 3,
-                "number_of_replicas": 2,
-                # Required for vector search.
-                "knn": True,
-                "knn.algo_param.ef_search": EF_SEARCH,
-            }
-        }
-
-    @staticmethod
-    def get_index_settings_for_aws_managed_opensearch_mt_cloud() -> dict[str, Any]:
-        """
-        Settings for AWS-managed OpenSearch in multi-tenant cloud.
-
-        324 shards very roughly targets a storage load of ~30Gb per shard, which
-        according to AWS OpenSearch documentation is within a good target range.
-
-        As documented above we need 2 replicas for a total of 3 copies of the
-        data because the cluster is configured with 3-AZ awareness.
-        """
-        return {
-            "index": {
-                "number_of_shards": 324,
-                "number_of_replicas": 2,
-                # Required for vector search.
-                "knn": True,
-                "knn.algo_param.ef_search": EF_SEARCH,
-            }
-        }
-
     @staticmethod
     def get_index_settings_based_on_environment() -> dict[str, Any]:
         """
         Returns the index settings based on the environment.
         """
         if USING_AWS_MANAGED_OPENSEARCH:
+            # NOTE: The number of data copies, including the primary (not a
+            # replica) copy, must be divisible by the number of AZs.
             if MULTI_TENANT:
-                return (
-                    DocumentSchema.get_index_settings_for_aws_managed_opensearch_mt_cloud()
-                )
+                number_of_shards = 324
+                number_of_replicas = 2
             else:
-                return (
-                    DocumentSchema.get_index_settings_for_aws_managed_opensearch_st_dev()
-                )
+                number_of_shards = 3
+                number_of_replicas = 2
         else:
-            return DocumentSchema.get_index_settings()
+            number_of_shards = 1
+            number_of_replicas = 1
+
+        if OPENSEARCH_INDEX_NUM_SHARDS is not None:
+            number_of_shards = OPENSEARCH_INDEX_NUM_SHARDS
+        if OPENSEARCH_INDEX_NUM_REPLICAS is not None:
+            number_of_replicas = OPENSEARCH_INDEX_NUM_REPLICAS
+
+        return {
+            "index": {
+                "number_of_shards": number_of_shards,
+                "number_of_replicas": number_of_replicas,
+                # Required for vector search.
+                "knn": True,
+                "knn.algo_param.ef_search": EF_SEARCH,
+            }
+        }

backend/onyx/document_index/opensearch/search.py

@@ -7,16 +7,28 @@ from uuid import UUID
 
 from onyx.configs.app_configs import DEFAULT_OPENSEARCH_QUERY_TIMEOUT_S
 from onyx.configs.app_configs import OPENSEARCH_EXPLAIN_ENABLED
+from onyx.configs.app_configs import OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED
 from onyx.configs.app_configs import OPENSEARCH_PROFILING_DISABLED
 from onyx.configs.constants import DocumentSource
 from onyx.configs.constants import INDEX_SEPARATOR
 from onyx.context.search.models import IndexFilters
 from onyx.context.search.models import Tag
 from onyx.document_index.interfaces_new import TenantState
+from onyx.document_index.opensearch.constants import ASSUMED_DOCUMENT_AGE_DAYS
 from onyx.document_index.opensearch.constants import (
-    DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES,
+    DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES,
 )
-from onyx.document_index.opensearch.constants import HYBRID_SEARCH_NORMALIZATION_WEIGHTS
+from onyx.document_index.opensearch.constants import (
+    DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW,
+)
+from onyx.document_index.opensearch.constants import (
+    HYBRID_SEARCH_NORMALIZATION_PIPELINE,
+)
+from onyx.document_index.opensearch.constants import (
+    HYBRID_SEARCH_SUBQUERY_CONFIGURATION,
+)
+from onyx.document_index.opensearch.constants import HybridSearchNormalizationPipeline
+from onyx.document_index.opensearch.constants import HybridSearchSubqueryConfiguration
 from onyx.document_index.opensearch.schema import ACCESS_CONTROL_LIST_FIELD_NAME
 from onyx.document_index.opensearch.schema import ANCESTOR_HIERARCHY_NODE_IDS_FIELD_NAME
 from onyx.document_index.opensearch.schema import CHUNK_INDEX_FIELD_NAME
@@ -43,49 +55,113 @@ from onyx.document_index.opensearch.schema import USER_PROJECTS_FIELD_NAME
 
 # TODO(andrei): Turn all magic dictionaries to pydantic models.
 
-MIN_MAX_NORMALIZATION_PIPELINE_NAME = "normalization_pipeline_min_max"
-MIN_MAX_NORMALIZATION_PIPELINE_CONFIG: dict[str, Any] = {
-    "description": "Normalization for keyword and vector scores using min-max",
-    "phase_results_processors": [
-        {
-            # https://docs.opensearch.org/latest/search-plugins/search-pipelines/normalization-processor/
-            "normalization-processor": {
-                "normalization": {"technique": "min_max"},
-                "combination": {
-                    "technique": "arithmetic_mean",
-                    "parameters": {"weights": HYBRID_SEARCH_NORMALIZATION_WEIGHTS},
-                },
+
+def _get_hybrid_search_normalization_weights() -> list[float]:
+    if (
+        HYBRID_SEARCH_SUBQUERY_CONFIGURATION
+        is HybridSearchSubqueryConfiguration.TITLE_VECTOR_CONTENT_VECTOR_TITLE_CONTENT_COMBINED_KEYWORD
+    ):
+        # Since the titles are included in the contents, the embedding matches
+        # are heavily downweighted as they act as a boost rather than an
+        # independent scoring component.
+        search_title_vector_weight = 0.1
+        search_content_vector_weight = 0.45
+        # Single keyword weight for both title and content (merged from former
+        # title keyword + content keyword).
+        search_keyword_weight = 0.45
+
+        # NOTE: It is critical that the order of these weights matches the order
+        # of the sub-queries in the hybrid search.
+        hybrid_search_normalization_weights = [
+            search_title_vector_weight,
+            search_content_vector_weight,
+            search_keyword_weight,
+        ]
+    elif (
+        HYBRID_SEARCH_SUBQUERY_CONFIGURATION
+        is HybridSearchSubqueryConfiguration.CONTENT_VECTOR_TITLE_CONTENT_COMBINED_KEYWORD
+    ):
+        search_content_vector_weight = 0.5
+        # Single keyword weight for both title and content (merged from former
+        # title keyword + content keyword).
+        search_keyword_weight = 0.5
+
+        # NOTE: It is critical that the order of these weights matches the order
+        # of the sub-queries in the hybrid search.
+        hybrid_search_normalization_weights = [
+            search_content_vector_weight,
+            search_keyword_weight,
+        ]
+    else:
+        raise ValueError(
+            f"Bug: Unhandled hybrid search subquery configuration: {HYBRID_SEARCH_SUBQUERY_CONFIGURATION}."
+        )
+
+    assert (
+        sum(hybrid_search_normalization_weights) == 1.0
+    ), "Bug: Hybrid search normalization weights do not sum to 1.0."
+
+    return hybrid_search_normalization_weights
+
+
+def get_min_max_normalization_pipeline_name_and_config() -> tuple[str, dict[str, Any]]:
+    min_max_normalization_pipeline_name = "normalization_pipeline_min_max"
+    min_max_normalization_pipeline_config: dict[str, Any] = {
+        "description": "Normalization for keyword and vector scores using min-max",
+        "phase_results_processors": [
+            {
+                # https://docs.opensearch.org/latest/search-plugins/search-pipelines/normalization-processor/
+                "normalization-processor": {
+                    "normalization": {"technique": "min_max"},
+                    "combination": {
+                        "technique": "arithmetic_mean",
+                        "parameters": {
+                            "weights": _get_hybrid_search_normalization_weights()
+                        },
+                    },
+                }
             }
-        }
-    ],
-}
+        ],
+    }
+    return min_max_normalization_pipeline_name, min_max_normalization_pipeline_config
 
-ZSCORE_NORMALIZATION_PIPELINE_NAME = "normalization_pipeline_zscore"
-ZSCORE_NORMALIZATION_PIPELINE_CONFIG: dict[str, Any] = {
-    "description": "Normalization for keyword and vector scores using z-score",
-    "phase_results_processors": [
-        {
-            # https://docs.opensearch.org/latest/search-plugins/search-pipelines/normalization-processor/
-            "normalization-processor": {
-                "normalization": {"technique": "z_score"},
-                "combination": {
-                    "technique": "arithmetic_mean",
-                    "parameters": {"weights": HYBRID_SEARCH_NORMALIZATION_WEIGHTS},
-                },
+
+def get_zscore_normalization_pipeline_name_and_config() -> tuple[str, dict[str, Any]]:
+    zscore_normalization_pipeline_name = "normalization_pipeline_zscore"
+    zscore_normalization_pipeline_config: dict[str, Any] = {
+        "description": "Normalization for keyword and vector scores using z-score",
+        "phase_results_processors": [
+            {
+                # https://docs.opensearch.org/latest/search-plugins/search-pipelines/normalization-processor/
+                "normalization-processor": {
+                    "normalization": {"technique": "z_score"},
+                    "combination": {
+                        "technique": "arithmetic_mean",
+                        "parameters": {
+                            "weights": _get_hybrid_search_normalization_weights()
+                        },
+                    },
+                }
             }
-        }
-    ],
-}
+        ],
+    }
+    return zscore_normalization_pipeline_name, zscore_normalization_pipeline_config
 
 
-# By default OpenSearch will only return a maximum of this many results in a
-# given search. This value is configurable in the index settings.
-DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW = 10_000
-
-# For documents which do not have a value for LAST_UPDATED_FIELD_NAME, we assume
-# that the document was last updated this many days ago for the purpose of time
-# cutoff filtering during retrieval.
-ASSUMED_DOCUMENT_AGE_DAYS = 90
+def get_normalization_pipeline_name_and_config() -> tuple[str, dict[str, Any]]:
+    if (
+        HYBRID_SEARCH_NORMALIZATION_PIPELINE
+        is HybridSearchNormalizationPipeline.MIN_MAX
+    ):
+        return get_min_max_normalization_pipeline_name_and_config()
+    elif (
+        HYBRID_SEARCH_NORMALIZATION_PIPELINE is HybridSearchNormalizationPipeline.ZSCORE
+    ):
+        return get_zscore_normalization_pipeline_name_and_config()
+    else:
+        raise ValueError(
+            f"Bug: Unhandled hybrid search normalization pipeline: {HYBRID_SEARCH_NORMALIZATION_PIPELINE}."
+        )
 
 
 class DocumentQuery:
@@ -160,9 +236,17 @@ class DocumentQuery:
             # returning some number of results less than the index max allowed
             # return size.
             "size": DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW,
-            "_source": get_full_document,
+            # By default exclude retrieving the vector fields in order to save
+            # on retrieval cost as we don't need them upstream.
+            "_source": {
+                "excludes": [TITLE_VECTOR_FIELD_NAME, CONTENT_VECTOR_FIELD_NAME]
+            },
             "timeout": f"{DEFAULT_OPENSEARCH_QUERY_TIMEOUT_S}s",
         }
+        if not get_full_document:
+            # If we explicitly do not want the underlying document, we will only
+            # retrieve IDs.
+            final_get_ids_query["_source"] = False
         if not OPENSEARCH_PROFILING_DISABLED:
             final_get_ids_query["profile"] = True
 
@@ -257,7 +341,7 @@ class DocumentQuery:
 
         # TODO(andrei, yuhong): We can tune this more dynamically based on
         # num_hits.
-        max_results_per_subquery = DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES
+        max_results_per_subquery = DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES
 
         hybrid_search_subqueries = DocumentQuery._get_hybrid_search_subqueries(
             query_text, query_vector, vector_candidates=max_results_per_subquery
@@ -281,9 +365,6 @@ class DocumentQuery:
             attached_document_ids=index_filters.attached_document_ids,
             hierarchy_node_ids=index_filters.hierarchy_node_ids,
         )
-        match_highlights_configuration = (
-            DocumentQuery._get_match_highlights_configuration()
-        )
 
         # See https://docs.opensearch.org/latest/query-dsl/compound/hybrid/
         hybrid_search_query: dict[str, Any] = {
@@ -310,16 +391,183 @@ class DocumentQuery:
         final_hybrid_search_body: dict[str, Any] = {
             "query": hybrid_search_query,
             "size": num_hits,
-            "highlight": match_highlights_configuration,
             "timeout": f"{DEFAULT_OPENSEARCH_QUERY_TIMEOUT_S}s",
+            # Exclude retrieving the vector fields in order to save on
+            # retrieval cost as we don't need them upstream.
+            "_source": {
+                "excludes": [TITLE_VECTOR_FIELD_NAME, CONTENT_VECTOR_FIELD_NAME]
+            },
         }
 
-        # Explain is for scoring breakdowns.
+        if not OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED:
+            final_hybrid_search_body["highlight"] = (
+                DocumentQuery._get_match_highlights_configuration()
+            )
+
+        # Explain is for scoring breakdowns. Setting this significantly
+        # increases query latency.
         if OPENSEARCH_EXPLAIN_ENABLED:
             final_hybrid_search_body["explain"] = True
 
         return final_hybrid_search_body
 
+    @staticmethod
+    def get_keyword_search_query(
+        query_text: str,
+        num_hits: int,
+        tenant_state: TenantState,
+        index_filters: IndexFilters,
+        include_hidden: bool,
+    ) -> dict[str, Any]:
+        """Returns a final keyword search query.
+
+        This query can be directly supplied to the OpenSearch client.
+
+        Args:
+            query_text: The text to query for.
+            num_hits: The final number of hits to return.
+            tenant_state: Tenant state containing the tenant ID.
+            index_filters: Filters for the keyword search query.
+            include_hidden: Whether to include hidden documents.
+
+        Returns:
+            A dictionary representing the final keyword search query.
+        """
+        if num_hits > DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW:
+            raise ValueError(
+                f"Bug: num_hits ({num_hits}) is greater than the current maximum allowed "
+                f"result window ({DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW})."
+            )
+
+        keyword_search_filters = DocumentQuery._get_search_filters(
+            tenant_state=tenant_state,
+            include_hidden=include_hidden,
+            # TODO(andrei): We've done no filtering for PUBLIC_DOC_PAT up to
+            # now. This should not cause any issues but it can introduce
+            # redundant filters in queries that may affect performance.
+            access_control_list=index_filters.access_control_list,
+            source_types=index_filters.source_type or [],
+            tags=index_filters.tags or [],
+            document_sets=index_filters.document_set or [],
+            user_file_ids=index_filters.user_file_ids or [],
+            project_id=index_filters.project_id,
+            persona_id=index_filters.persona_id,
+            time_cutoff=index_filters.time_cutoff,
+            min_chunk_index=None,
+            max_chunk_index=None,
+            attached_document_ids=index_filters.attached_document_ids,
+            hierarchy_node_ids=index_filters.hierarchy_node_ids,
+        )
+
+        keyword_search_query = (
+            DocumentQuery._get_title_content_combined_keyword_search_query(
+                query_text, search_filters=keyword_search_filters
+            )
+        )
+
+        final_keyword_search_query: dict[str, Any] = {
+            "query": keyword_search_query,
+            "size": num_hits,
+            "timeout": f"{DEFAULT_OPENSEARCH_QUERY_TIMEOUT_S}s",
+            # Exclude retrieving the vector fields in order to save on
+            # retrieval cost as we don't need them upstream.
+            "_source": {
+                "excludes": [TITLE_VECTOR_FIELD_NAME, CONTENT_VECTOR_FIELD_NAME]
+            },
+        }
+
+        if not OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED:
+            final_keyword_search_query["highlight"] = (
+                DocumentQuery._get_match_highlights_configuration()
+            )
+
+        if not OPENSEARCH_PROFILING_DISABLED:
+            final_keyword_search_query["profile"] = True
+
+        # Explain is for scoring breakdowns. Setting this significantly
+        # increases query latency.
+        if OPENSEARCH_EXPLAIN_ENABLED:
+            final_keyword_search_query["explain"] = True
+
+        return final_keyword_search_query
+
+    @staticmethod
+    def get_semantic_search_query(
+        query_embedding: list[float],
+        num_hits: int,
+        tenant_state: TenantState,
+        index_filters: IndexFilters,
+        include_hidden: bool,
+    ) -> dict[str, Any]:
+        """Returns a final semantic search query.
+
+        This query can be directly supplied to the OpenSearch client.
+
+        Args:
+            query_embedding: The vector embedding of the text to query for.
+            num_hits: The final number of hits to return.
+            tenant_state: Tenant state containing the tenant ID.
+            index_filters: Filters for the semantic search query.
+            include_hidden: Whether to include hidden documents.
+
+        Returns:
+            A dictionary representing the final semantic search query.
+        """
+        if num_hits > DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW:
+            raise ValueError(
+                f"Bug: num_hits ({num_hits}) is greater than the current maximum allowed "
+                f"result window ({DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW})."
+            )
+
+        semantic_search_filters = DocumentQuery._get_search_filters(
+            tenant_state=tenant_state,
+            include_hidden=include_hidden,
+            # TODO(andrei): We've done no filtering for PUBLIC_DOC_PAT up to
+            # now. This should not cause any issues but it can introduce
+            # redundant filters in queries that may affect performance.
+            access_control_list=index_filters.access_control_list,
+            source_types=index_filters.source_type or [],
+            tags=index_filters.tags or [],
+            document_sets=index_filters.document_set or [],
+            user_file_ids=index_filters.user_file_ids or [],
+            project_id=index_filters.project_id,
+            persona_id=index_filters.persona_id,
+            time_cutoff=index_filters.time_cutoff,
+            min_chunk_index=None,
+            max_chunk_index=None,
+            attached_document_ids=index_filters.attached_document_ids,
+            hierarchy_node_ids=index_filters.hierarchy_node_ids,
+        )
+
+        semantic_search_query = (
+            DocumentQuery._get_content_vector_similarity_search_query(
+                query_embedding,
+                vector_candidates=num_hits,
+                search_filters=semantic_search_filters,
+            )
+        )
+
+        final_semantic_search_query: dict[str, Any] = {
+            "query": semantic_search_query,
+            "size": num_hits,
+            "timeout": f"{DEFAULT_OPENSEARCH_QUERY_TIMEOUT_S}s",
+            # Exclude retrieving the vector fields in order to save on
+            # retrieval cost as we don't need them upstream.
+            "_source": {
+                "excludes": [TITLE_VECTOR_FIELD_NAME, CONTENT_VECTOR_FIELD_NAME]
+            },
+        }
+
+        if not OPENSEARCH_PROFILING_DISABLED:
+            final_semantic_search_query["profile"] = True
+
+        # Explain is for scoring breakdowns. Setting this significantly
+        # increases query latency.
+        if OPENSEARCH_EXPLAIN_ENABLED:
+            final_semantic_search_query["explain"] = True
+
+        return final_semantic_search_query
+
     @staticmethod
     def get_random_search_query(
         tenant_state: TenantState,
@@ -371,6 +619,11 @@ class DocumentQuery:
             },
             "size": num_to_retrieve,
             "timeout": f"{DEFAULT_OPENSEARCH_QUERY_TIMEOUT_S}s",
+            # Exclude retrieving the vector fields in order to save on
+            # retrieval cost as we don't need them upstream.
+            "_source": {
+                "excludes": [TITLE_VECTOR_FIELD_NAME, CONTENT_VECTOR_FIELD_NAME]
+            },
         }
         if not OPENSEARCH_PROFILING_DISABLED:
             final_random_search_query["profile"] = True
@@ -385,7 +638,7 @@ class DocumentQuery:
         # search. This is higher than the number of results because the scoring
         # is hybrid. For a detailed breakdown, see where the default value is
         # set.
-        vector_candidates: int = DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES,
+        vector_candidates: int = DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES,
     ) -> list[dict[str, Any]]:
         """Returns subqueries for hybrid search.
 
@@ -395,20 +648,18 @@ class DocumentQuery:
         The return of this function is not sufficient to be directly supplied to
         the OpenSearch client. See get_hybrid_search_query.
 
-        Matches:
-          - Title vector
-          - Content vector
-          - Keyword (title + content, match and phrase)
-
         Normalization is not performed here.
         The weights of each of these subqueries should be configured in a search
         pipeline.
 
+        The exact subqueries executed depend on the
+        HYBRID_SEARCH_SUBQUERY_CONFIGURATION setting.
+
         NOTE: For OpenSearch, 5 is the maximum number of query clauses allowed
         in a single hybrid query. Source:
         https://docs.opensearch.org/latest/query-dsl/compound/hybrid/
 
-        NOTE: Each query is independent during the search phase, there is no
+        NOTE: Each query is independent during the search phase; there is no
         backfilling of scores for missing query components. What this means is
         that if a document was a good vector match but did not show up for
         keyword, it gets a score of 0 for the keyword component of the hybrid
@@ -437,74 +688,133 @@ class DocumentQuery:
                 similarity search.
         """
         # Build sub-queries for hybrid search. Order must match normalization
-        # pipeline weights: title vector, content vector, keyword (title + content).
-        hybrid_search_queries: list[dict[str, Any]] = [
-            # 1. Title vector search
-            {
-                "knn": {
-                    TITLE_VECTOR_FIELD_NAME: {
-                        "vector": query_vector,
-                        "k": vector_candidates,
-                    }
-                }
-            },
-            # 2. Content vector search
-            {
-                "knn": {
-                    CONTENT_VECTOR_FIELD_NAME: {
-                        "vector": query_vector,
-                        "k": vector_candidates,
-                    }
-                }
-            },
-            # 3. Keyword (title + content) match and phrase search.
-            {
-                "bool": {
-                    "should": [
-                        {
-                            "match": {
-                                TITLE_FIELD_NAME: {
-                                    "query": query_text,
-                                    "operator": "or",
-                                    # The title fields are strongly discounted as they are included in the content.
-                                    # It just acts as a minor boost
-                                    "boost": 0.1,
-                                }
-                            }
-                        },
-                        {
-                            "match_phrase": {
-                                TITLE_FIELD_NAME: {
-                                    "query": query_text,
-                                    "slop": 1,
-                                    "boost": 0.2,
-                                }
-                            }
-                        },
-                        {
-                            "match": {
-                                CONTENT_FIELD_NAME: {
-                                    "query": query_text,
-                                    "operator": "or",
-                                    "boost": 1.0,
-                                }
-                            }
-                        },
-                        {
-                            "match_phrase": {
-                                CONTENT_FIELD_NAME: {
-                                    "query": query_text,
-                                    "slop": 1,
-                                    "boost": 1.5,
-                                }
-                            }
-                        },
-                    ]
-                }
-            },
-        ]
+        # pipeline weights.
+        if (
+            HYBRID_SEARCH_SUBQUERY_CONFIGURATION
+            is HybridSearchSubqueryConfiguration.TITLE_VECTOR_CONTENT_VECTOR_TITLE_CONTENT_COMBINED_KEYWORD
+        ):
+            return [
+                DocumentQuery._get_title_vector_similarity_search_query(
+                    query_vector, vector_candidates
+                ),
+                DocumentQuery._get_content_vector_similarity_search_query(
+                    query_vector, vector_candidates
+                ),
+                DocumentQuery._get_title_content_combined_keyword_search_query(
+                    query_text
+                ),
+            ]
+        elif (
+            HYBRID_SEARCH_SUBQUERY_CONFIGURATION
+            is HybridSearchSubqueryConfiguration.CONTENT_VECTOR_TITLE_CONTENT_COMBINED_KEYWORD
+        ):
+            return [
+                DocumentQuery._get_content_vector_similarity_search_query(
+                    query_vector, vector_candidates
+                ),
+                DocumentQuery._get_title_content_combined_keyword_search_query(
+                    query_text
+                ),
+            ]
+        else:
+            raise ValueError(
+                f"Bug: Unhandled hybrid search subquery configuration: {HYBRID_SEARCH_SUBQUERY_CONFIGURATION}"
+            )
 
-        return hybrid_search_queries
+    @staticmethod
+    def _get_title_vector_similarity_search_query(
+        query_vector: list[float],
+        vector_candidates: int = DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES,
+    ) -> dict[str, Any]:
+        return {
+            "knn": {
+                TITLE_VECTOR_FIELD_NAME: {
+                    "vector": query_vector,
+                    "k": vector_candidates,
+                }
+            }
+        }
+
+    @staticmethod
+    def _get_content_vector_similarity_search_query(
+        query_vector: list[float],
+        vector_candidates: int = DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES,
+        search_filters: list[dict[str, Any]] | None = None,
+    ) -> dict[str, Any]:
+        query = {
+            "knn": {
+                CONTENT_VECTOR_FIELD_NAME: {
+                    "vector": query_vector,
+                    "k": vector_candidates,
+                }
+            }
+        }
+
+        if search_filters is not None:
+            query["knn"][CONTENT_VECTOR_FIELD_NAME]["filter"] = {
+                "bool": {"filter": search_filters}
+            }
+
+        return query
+
+    @staticmethod
+    def _get_title_content_combined_keyword_search_query(
+        query_text: str,
+        search_filters: list[dict[str, Any]] | None = None,
+    ) -> dict[str, Any]:
+        query = {
+            "bool": {
+                "should": [
+                    {
+                        "match": {
+                            TITLE_FIELD_NAME: {
+                                "query": query_text,
+                                "operator": "or",
+                                # The title fields are strongly discounted as they are included in the content.
+                                # It just acts as a minor boost
+                                "boost": 0.1,
+                            }
+                        }
+                    },
+                    {
+                        "match_phrase": {
+                            TITLE_FIELD_NAME: {
+                                "query": query_text,
+                                "slop": 1,
+                                "boost": 0.2,
+                            }
+                        }
+                    },
+                    {
+                        "match": {
+                            CONTENT_FIELD_NAME: {
+                                "query": query_text,
+                                "operator": "or",
+                                "boost": 1.0,
+                            }
+                        }
+                    },
+                    {
+                        "match_phrase": {
+                            CONTENT_FIELD_NAME: {
+                                "query": query_text,
+                                "slop": 1,
+                                "boost": 1.5,
+                            }
+                        }
+                    },
+                ],
+                # Ensure at least one term from the query is present in the
+                # document. This defaults to 1, unless a filter or must clause
+                # is supplied, in which case it defaults to 0.
+                "minimum_should_match": 1,
+            }
+        }
+
+        if search_filters is not None:
+            query["bool"]["filter"] = search_filters
+
+        return query
 
     @staticmethod
     def _get_search_filters(

backend/onyx/error_handling/error_codes.py

@@ -44,6 +44,7 @@ class OnyxErrorCode(Enum):
     VALIDATION_ERROR = ("VALIDATION_ERROR", 400)
     INVALID_INPUT = ("INVALID_INPUT", 400)
     MISSING_REQUIRED_FIELD = ("MISSING_REQUIRED_FIELD", 400)
+    QUERY_REJECTED = ("QUERY_REJECTED", 400)
 
     # ------------------------------------------------------------------
     # Not Found (404)
@@ -88,6 +89,7 @@ class OnyxErrorCode(Enum):
     SERVICE_UNAVAILABLE = ("SERVICE_UNAVAILABLE", 503)
     BAD_GATEWAY = ("BAD_GATEWAY", 502)
     LLM_PROVIDER_ERROR = ("LLM_PROVIDER_ERROR", 502)
+    HOOK_EXECUTION_FAILED = ("HOOK_EXECUTION_FAILED", 502)
     GATEWAY_TIMEOUT = ("GATEWAY_TIMEOUT", 504)
 
     def __init__(self, code: str, status_code: int) -> None:

backend/onyx/file_processing/image_summarization.py

@@ -88,9 +88,13 @@ def summarize_image_with_error_handling(
     try:
         return summarize_image_pipeline(llm, image_data, user_prompt, system_prompt)
     except UnsupportedImageFormatError:
+        magic_hex = image_data[:8].hex() if image_data else "empty"
         logger.info(
-            "Skipping image summarization due to unsupported MIME type for %s",
+            "Skipping image summarization due to unsupported MIME type "
+            "for %s (magic_bytes=%s, size=%d bytes)",
             context_name,
+            magic_hex,
+            len(image_data),
         )
         return None
 
@@ -134,9 +138,23 @@ def _summarize_image(
         return summary
 
     except Exception as e:
-        error_msg = f"Summarization failed. Messages: {messages}"
-        error_msg = error_msg[:1024]
-        raise ValueError(error_msg) from e
+        # Extract structured details from LiteLLM exceptions when available,
+        # rather than dumping the full messages payload (which contains base64
+        # image data and produces enormous, unreadable error logs).
+        str_e = str(e)
+        if len(str_e) > 512:
+            str_e = str_e[:512] + "... (truncated)"
+        parts = [f"Summarization failed: {type(e).__name__}: {str_e}"]
+        status_code = getattr(e, "status_code", None)
+        llm_provider = getattr(e, "llm_provider", None)
+        model = getattr(e, "model", None)
+        if status_code is not None:
+            parts.append(f"status_code={status_code}")
+        if llm_provider is not None:
+            parts.append(f"llm_provider={llm_provider}")
+        if model is not None:
+            parts.append(f"model={model}")
+        raise ValueError(" | ".join(parts)) from e
 
 
 def _encode_image_for_llm_prompt(image_data: bytes) -> str:

backend/onyx/hooks/executor.py

@@ -0,0 +1,387 @@
+"""Hook executor — calls a customer's external HTTP endpoint for a given hook point.
+
+Usage (Celery tasks and FastAPI handlers):
+    result = execute_hook(
+        db_session=db_session,
+        hook_point=HookPoint.QUERY_PROCESSING,
+        payload={"query": "...", "user_email": "...", "chat_session_id": "..."},
+    )
+
+    if isinstance(result, HookSkipped):
+        # no active hook configured — continue with original behavior
+        ...
+    elif isinstance(result, HookSoftFailed):
+        # hook failed but fail strategy is SOFT — continue with original behavior
+        ...
+    else:
+        # result is a validated Pydantic model instance (spec.response_model)
+        ...
+
+is_reachable update policy
+--------------------------
+``is_reachable`` on the Hook row is updated selectively — only when the outcome
+carries meaningful signal about physical reachability:
+
+  NetworkError (DNS, connection refused)  → False  (cannot reach the server)
+  HTTP 401 / 403                          → False  (api_key revoked or invalid)
+  TimeoutException                        → None   (server may be slow, skip write)
+  Other HTTP errors (4xx / 5xx)           → None   (server responded, skip write)
+  Unknown exception                       → None   (no signal, skip write)
+  Non-JSON / non-dict response            → None   (server responded, skip write)
+  Success (2xx, valid dict)               → True   (confirmed reachable)
+
+None means "leave the current value unchanged" — no DB round-trip is made.
+
+DB session design
+-----------------
+The executor uses three sessions:
+
+  1. Caller's session (db_session) — used only for the hook lookup read. All
+     needed fields are extracted from the Hook object before the HTTP call, so
+     the caller's session is not held open during the external HTTP request.
+
+  2. Log session — a separate short-lived session opened after the HTTP call
+     completes to write the HookExecutionLog row on failure. Success runs are
+     not recorded. Committed independently of everything else.
+
+  3. Reachable session — a second short-lived session to update is_reachable on
+     the Hook. Kept separate from the log session so a concurrent hook deletion
+     (which causes update_hook__no_commit to raise OnyxError(NOT_FOUND)) cannot
+     prevent the execution log from being written. This update is best-effort.
+"""
+
+import json
+import time
+from typing import Any
+
+import httpx
+from pydantic import BaseModel
+from pydantic import ValidationError
+from sqlalchemy.orm import Session
+
+from onyx.db.engine.sql_engine import get_session_with_current_tenant
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.db.hook import create_hook_execution_log__no_commit
+from onyx.db.hook import get_non_deleted_hook_by_hook_point
+from onyx.db.hook import update_hook__no_commit
+from onyx.db.models import Hook
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
+from onyx.hooks.registry import get_hook_point_spec
+from onyx.hooks.utils import HOOKS_AVAILABLE
+from onyx.utils.logger import setup_logger
+
+logger = setup_logger()
+
+
+class HookSkipped:
+    """No active hook configured for this hook point."""
+
+
+class HookSoftFailed:
+    """Hook was called but failed with SOFT fail strategy — continuing."""
+
+
+# ---------------------------------------------------------------------------
+# Private helpers
+# ---------------------------------------------------------------------------
+
+
+class _HttpOutcome(BaseModel):
+    """Structured result of an HTTP hook call, returned by _process_response."""
+
+    is_success: bool
+    updated_is_reachable: (
+        bool | None
+    )  # True/False = write to DB, None = unchanged (skip write)
+    status_code: int | None
+    error_message: str | None
+    response_payload: dict[str, Any] | None
+
+
+def _lookup_hook(
+    db_session: Session,
+    hook_point: HookPoint,
+) -> Hook | HookSkipped:
+    """Return the active Hook or HookSkipped if hooks are unavailable/unconfigured.
+
+    No HTTP call is made and no DB writes are performed for any HookSkipped path.
+    There is nothing to log and no reachability information to update.
+    """
+    if not HOOKS_AVAILABLE:
+        return HookSkipped()
+    hook = get_non_deleted_hook_by_hook_point(
+        db_session=db_session, hook_point=hook_point
+    )
+    if hook is None or not hook.is_active:
+        return HookSkipped()
+    if not hook.endpoint_url:
+        return HookSkipped()
+    return hook
+
+
+def _process_response(
+    *,
+    response: httpx.Response | None,
+    exc: Exception | None,
+    timeout: float,
+) -> _HttpOutcome:
+    """Process the result of an HTTP call and return a structured outcome.
+
+    Called after the client.post() try/except. If post() raised, exc is set and
+    response is None. Otherwise response is set and exc is None. Handles
+    raise_for_status(), JSON decoding, and the dict shape check.
+    """
+    if exc is not None:
+        if isinstance(exc, httpx.NetworkError):
+            msg = f"Hook network error (endpoint unreachable): {exc}"
+            logger.warning(msg, exc_info=exc)
+            return _HttpOutcome(
+                is_success=False,
+                updated_is_reachable=False,
+                status_code=None,
+                error_message=msg,
+                response_payload=None,
+            )
+        if isinstance(exc, httpx.TimeoutException):
+            msg = f"Hook timed out after {timeout}s: {exc}"
+            logger.warning(msg, exc_info=exc)
+            return _HttpOutcome(
+                is_success=False,
+                updated_is_reachable=None,  # timeout doesn't indicate unreachability
+                status_code=None,
+                error_message=msg,
+                response_payload=None,
+            )
+        msg = f"Hook call failed: {exc}"
+        logger.exception(msg, exc_info=exc)
+        return _HttpOutcome(
+            is_success=False,
+            updated_is_reachable=None,  # unknown error — don't make assumptions
+            status_code=None,
+            error_message=msg,
+            response_payload=None,
+        )
+
+    if response is None:
+        raise ValueError(
+            "exactly one of response or exc must be non-None; both are None"
+        )
+    status_code = response.status_code
+
+    try:
+        response.raise_for_status()
+    except httpx.HTTPStatusError as e:
+        msg = f"Hook returned HTTP {e.response.status_code}: {e.response.text}"
+        logger.warning(msg, exc_info=e)
+        # 401/403 means the api_key has been revoked or is invalid — mark unreachable
+        # so the operator knows to update it. All other HTTP errors keep is_reachable
+        # as-is (server is up, the request just failed for application reasons).
+        auth_failed = e.response.status_code in (401, 403)
+        return _HttpOutcome(
+            is_success=False,
+            updated_is_reachable=False if auth_failed else None,
+            status_code=status_code,
+            error_message=msg,
+            response_payload=None,
+        )
+
+    try:
+        response_payload = response.json()
+    except (json.JSONDecodeError, httpx.DecodingError) as e:
+        msg = f"Hook returned non-JSON response: {e}"
+        logger.warning(msg, exc_info=e)
+        return _HttpOutcome(
+            is_success=False,
+            updated_is_reachable=None,  # server responded — reachability unchanged
+            status_code=status_code,
+            error_message=msg,
+            response_payload=None,
+        )
+
+    if not isinstance(response_payload, dict):
+        msg = f"Hook returned non-dict JSON (got {type(response_payload).__name__})"
+        logger.warning(msg)
+        return _HttpOutcome(
+            is_success=False,
+            updated_is_reachable=None,  # server responded — reachability unchanged
+            status_code=status_code,
+            error_message=msg,
+            response_payload=None,
+        )
+
+    return _HttpOutcome(
+        is_success=True,
+        updated_is_reachable=True,
+        status_code=status_code,
+        error_message=None,
+        response_payload=response_payload,
+    )
+
+
+def _persist_result(
+    *,
+    hook_id: int,
+    outcome: _HttpOutcome,
+    duration_ms: int,
+) -> None:
+    """Write the execution log on failure and optionally update is_reachable, each
+    in its own session so a failure in one does not affect the other."""
+    # Only write the execution log on failure — success runs are not recorded.
+    # Must not be skipped if the is_reachable update fails (e.g. hook concurrently
+    # deleted between the initial lookup and here).
+    if not outcome.is_success:
+        try:
+            with get_session_with_current_tenant() as log_session:
+                create_hook_execution_log__no_commit(
+                    db_session=log_session,
+                    hook_id=hook_id,
+                    is_success=False,
+                    error_message=outcome.error_message,
+                    status_code=outcome.status_code,
+                    duration_ms=duration_ms,
+                )
+                log_session.commit()
+        except Exception:
+            logger.exception(
+                f"Failed to persist hook execution log for hook_id={hook_id}"
+            )
+
+    # Update is_reachable separately — best-effort, non-critical.
+    # None means the value is unchanged (set by the caller to skip the no-op write).
+    # update_hook__no_commit can raise OnyxError(NOT_FOUND) if the hook was
+    # concurrently deleted, so keep this isolated from the log write above.
+    if outcome.updated_is_reachable is not None:
+        try:
+            with get_session_with_current_tenant() as reachable_session:
+                update_hook__no_commit(
+                    db_session=reachable_session,
+                    hook_id=hook_id,
+                    is_reachable=outcome.updated_is_reachable,
+                )
+                reachable_session.commit()
+        except Exception:
+            logger.warning(f"Failed to update is_reachable for hook_id={hook_id}")
+
+
+# ---------------------------------------------------------------------------
+# Public API
+# ---------------------------------------------------------------------------
+
+
+def _execute_hook_inner(
+    hook: Hook,
+    payload: dict[str, Any],
+) -> BaseModel | HookSoftFailed:
+    """Make the HTTP call, validate the response, and return a typed model.
+
+    Raises OnyxError on HARD failure. Returns HookSoftFailed on SOFT failure.
+    """
+    timeout = hook.timeout_seconds
+    hook_id = hook.id
+    fail_strategy = hook.fail_strategy
+    endpoint_url = hook.endpoint_url
+    current_is_reachable: bool | None = hook.is_reachable
+    hook_point = hook.hook_point  # extract before HTTP call per design intent
+
+    if not endpoint_url:
+        raise ValueError(
+            f"hook_id={hook_id} is active but has no endpoint_url — "
+            "active hooks without an endpoint_url must be rejected by _lookup_hook"
+        )
+
+    start = time.monotonic()
+    response: httpx.Response | None = None
+    exc: Exception | None = None
+    try:
+        api_key: str | None = (
+            hook.api_key.get_value(apply_mask=False) if hook.api_key else None
+        )
+        headers: dict[str, str] = {"Content-Type": "application/json"}
+        if api_key:
+            headers["Authorization"] = f"Bearer {api_key}"
+        with httpx.Client(
+            timeout=timeout, follow_redirects=False
+        ) as client:  # SSRF guard: never follow redirects
+            response = client.post(endpoint_url, json=payload, headers=headers)
+    except Exception as e:
+        exc = e
+    duration_ms = int((time.monotonic() - start) * 1000)
+
+    outcome = _process_response(response=response, exc=exc, timeout=timeout)
+
+    # Validate the response payload against the spec's response_model.
+    # A validation failure downgrades the outcome to a failure so it is logged,
+    # is_reachable is left unchanged (server responded — just a bad payload),
+    # and fail_strategy is respected below.
+    validated_model: BaseModel | None = None
+    if outcome.is_success and outcome.response_payload is not None:
+        spec = get_hook_point_spec(hook_point)
+        try:
+            validated_model = spec.response_model.model_validate(
+                outcome.response_payload
+            )
+        except ValidationError as e:
+            msg = f"Hook response failed validation against {spec.response_model.__name__}: {e}"
+            outcome = _HttpOutcome(
+                is_success=False,
+                updated_is_reachable=None,  # server responded — reachability unchanged
+                status_code=outcome.status_code,
+                error_message=msg,
+                response_payload=None,
+            )
+
+    # Skip the is_reachable write when the value would not change — avoids a
+    # no-op DB round-trip on every call when the hook is already in the expected state.
+    if outcome.updated_is_reachable == current_is_reachable:
+        outcome = outcome.model_copy(update={"updated_is_reachable": None})
+    _persist_result(hook_id=hook_id, outcome=outcome, duration_ms=duration_ms)
+
+    if not outcome.is_success:
+        if fail_strategy == HookFailStrategy.HARD:
+            raise OnyxError(
+                OnyxErrorCode.HOOK_EXECUTION_FAILED,
+                outcome.error_message or "Hook execution failed.",
+            )
+        logger.warning(
+            f"Hook execution failed (soft fail) for hook_id={hook_id}: {outcome.error_message}"
+        )
+        return HookSoftFailed()
+
+    if validated_model is None:
+        raise OnyxError(
+            OnyxErrorCode.INTERNAL_ERROR,
+            f"validated_model is None for successful hook call (hook_id={hook_id})",
+        )
+    return validated_model
+
+
+def execute_hook(
+    *,
+    db_session: Session,
+    hook_point: HookPoint,
+    payload: dict[str, Any],
+) -> BaseModel | HookSkipped | HookSoftFailed:
+    """Execute the hook for the given hook point synchronously.
+
+    Returns HookSkipped if no active hook is configured, HookSoftFailed if the
+    hook failed with SOFT fail strategy, or a validated response model on success.
+    Raises OnyxError on HARD failure or if the hook is misconfigured.
+    """
+    hook = _lookup_hook(db_session, hook_point)
+    if isinstance(hook, HookSkipped):
+        return hook
+
+    fail_strategy = hook.fail_strategy
+    hook_id = hook.id
+
+    try:
+        return _execute_hook_inner(hook, payload)
+    except Exception:
+        if fail_strategy == HookFailStrategy.SOFT:
+            logger.exception(
+                f"Unexpected error in hook execution (soft fail) for hook_id={hook_id}"
+            )
+            return HookSoftFailed()
+        raise

backend/onyx/hooks/models.py

@@ -0,0 +1,121 @@
+from datetime import datetime
+from enum import Enum
+from typing import Annotated
+from typing import Any
+
+from pydantic import BaseModel
+from pydantic import Field
+from pydantic import field_validator
+from pydantic import model_validator
+from pydantic import SecretStr
+
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+
+NonEmptySecretStr = Annotated[SecretStr, Field(min_length=1)]
+
+
+# ---------------------------------------------------------------------------
+# Request models
+# ---------------------------------------------------------------------------
+
+
+class HookCreateRequest(BaseModel):
+    name: str = Field(min_length=1)
+    hook_point: HookPoint
+    endpoint_url: str = Field(min_length=1)
+    api_key: NonEmptySecretStr | None = None
+    fail_strategy: HookFailStrategy | None = None  # if None, uses HookPointSpec default
+    timeout_seconds: float | None = Field(
+        default=None, gt=0
+    )  # if None, uses HookPointSpec default
+
+    @field_validator("name", "endpoint_url")
+    @classmethod
+    def no_whitespace_only(cls, v: str) -> str:
+        if not v.strip():
+            raise ValueError("cannot be whitespace-only.")
+        return v
+
+
+class HookUpdateRequest(BaseModel):
+    name: str | None = None
+    endpoint_url: str | None = None
+    api_key: NonEmptySecretStr | None = None
+    fail_strategy: HookFailStrategy | None = None
+    timeout_seconds: float | None = Field(default=None, gt=0)
+
+    @model_validator(mode="after")
+    def require_at_least_one_field(self) -> "HookUpdateRequest":
+        if not self.model_fields_set:
+            raise ValueError("At least one field must be provided for an update.")
+        if "name" in self.model_fields_set and not (self.name or "").strip():
+            raise ValueError("name cannot be cleared.")
+        if (
+            "endpoint_url" in self.model_fields_set
+            and not (self.endpoint_url or "").strip()
+        ):
+            raise ValueError("endpoint_url cannot be cleared.")
+        if "fail_strategy" in self.model_fields_set and self.fail_strategy is None:
+            raise ValueError(
+                "fail_strategy cannot be null; omit the field to leave it unchanged."
+            )
+        if "timeout_seconds" in self.model_fields_set and self.timeout_seconds is None:
+            raise ValueError(
+                "timeout_seconds cannot be null; omit the field to leave it unchanged."
+            )
+        return self
+
+
+# ---------------------------------------------------------------------------
+# Response models
+# ---------------------------------------------------------------------------
+
+
+class HookPointMetaResponse(BaseModel):
+    hook_point: HookPoint
+    display_name: str
+    description: str
+    docs_url: str | None
+    input_schema: dict[str, Any]
+    output_schema: dict[str, Any]
+    default_timeout_seconds: float
+    default_fail_strategy: HookFailStrategy
+    fail_hard_description: str
+
+
+class HookResponse(BaseModel):
+    id: int
+    name: str
+    hook_point: HookPoint
+    # Nullable to match the DB column — endpoint_url is required on creation but
+    # future hook point types may not use an external endpoint (e.g. built-in handlers).
+    endpoint_url: str | None
+    fail_strategy: HookFailStrategy
+    timeout_seconds: float  # always resolved — None from request is replaced with spec default before DB write
+    is_active: bool
+    is_reachable: bool | None
+    creator_email: str | None
+    created_at: datetime
+    updated_at: datetime
+
+
+class HookValidateStatus(str, Enum):
+    passed = "passed"  # server responded (any status except 401/403)
+    auth_failed = "auth_failed"  # server responded with 401 or 403
+    timeout = (
+        "timeout"  # TCP connected, but read/write timed out (server exists but slow)
+    )
+    cannot_connect = "cannot_connect"  # could not connect to the server
+
+
+class HookValidateResponse(BaseModel):
+    status: HookValidateStatus
+    error_message: str | None = None
+
+
+class HookExecutionRecord(BaseModel):
+    error_message: str | None = None
+    status_code: int | None = None
+    duration_ms: int | None = None
+    created_at: datetime

backend/onyx/hooks/points/base.py

@@ -0,0 +1,74 @@
+from typing import Any
+from typing import ClassVar
+
+from pydantic import BaseModel
+
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+
+
+_REQUIRED_ATTRS = (
+    "hook_point",
+    "display_name",
+    "description",
+    "default_timeout_seconds",
+    "fail_hard_description",
+    "default_fail_strategy",
+    "payload_model",
+    "response_model",
+)
+
+
+class HookPointSpec:
+    """Static metadata and contract for a pipeline hook point.
+
+    Each concrete subclass represents exactly one hook point and is instantiated
+    once at startup, registered in onyx.hooks.registry._REGISTRY. Prefer
+    get_hook_point_spec() or get_all_specs() from the registry over direct
+    instantiation.
+
+    Each hook point is a concrete subclass of this class. Onyx engineers
+    own these definitions — customers never touch this code.
+
+    Subclasses must define all attributes as class-level constants.
+    payload_model and response_model must be Pydantic BaseModel subclasses;
+    input_schema and output_schema are derived from them automatically.
+    """
+
+    hook_point: HookPoint
+    display_name: str
+    description: str
+    default_timeout_seconds: float
+    fail_hard_description: str
+    default_fail_strategy: HookFailStrategy
+    docs_url: str | None = None
+
+    payload_model: ClassVar[type[BaseModel]]
+    response_model: ClassVar[type[BaseModel]]
+
+    # Computed once at class definition time from payload_model / response_model.
+    input_schema: ClassVar[dict[str, Any]]
+    output_schema: ClassVar[dict[str, Any]]
+
+    def __init_subclass__(cls, **kwargs: object) -> None:
+        """Enforce that every subclass declares all required class attributes.
+
+        Called automatically by Python whenever a class inherits from HookPointSpec.
+        Raises TypeError at import time if any required attribute is missing or if
+        payload_model / response_model are not Pydantic BaseModel subclasses.
+        input_schema and output_schema are derived automatically from the models.
+        """
+        super().__init_subclass__(**kwargs)
+        missing = [attr for attr in _REQUIRED_ATTRS if not hasattr(cls, attr)]
+        if missing:
+            raise TypeError(f"{cls.__name__} must define class attributes: {missing}")
+        for attr in ("payload_model", "response_model"):
+            val = getattr(cls, attr, None)
+            if val is None or not (
+                isinstance(val, type) and issubclass(val, BaseModel)
+            ):
+                raise TypeError(
+                    f"{cls.__name__}.{attr} must be a Pydantic BaseModel subclass, got {val!r}"
+                )
+        cls.input_schema = cls.payload_model.model_json_schema()
+        cls.output_schema = cls.response_model.model_json_schema()

backend/onyx/hooks/points/document_ingestion.py

@@ -0,0 +1,31 @@
+from pydantic import BaseModel
+
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.hooks.points.base import HookPointSpec
+
+
+# TODO(@Bo-Onyx): define payload and response fields
+class DocumentIngestionPayload(BaseModel):
+    pass
+
+
+class DocumentIngestionResponse(BaseModel):
+    pass
+
+
+class DocumentIngestionSpec(HookPointSpec):
+    """Hook point that runs during document ingestion.
+
+    # TODO(@Bo-Onyx): define call site, input/output schema, and timeout budget.
+    """
+
+    hook_point = HookPoint.DOCUMENT_INGESTION
+    display_name = "Document Ingestion"
+    description = "Runs during document ingestion. Allows filtering or transforming documents before indexing."
+    default_timeout_seconds = 30.0
+    fail_hard_description = "The document will not be indexed."
+    default_fail_strategy = HookFailStrategy.HARD
+
+    payload_model = DocumentIngestionPayload
+    response_model = DocumentIngestionResponse

backend/onyx/hooks/points/query_processing.py

@@ -0,0 +1,70 @@
+from pydantic import BaseModel
+from pydantic import ConfigDict
+from pydantic import Field
+
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.hooks.points.base import HookPointSpec
+
+
+class QueryProcessingPayload(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    query: str = Field(description="The raw query string exactly as the user typed it.")
+    user_email: str | None = Field(
+        description="Email of the user submitting the query, or null if unauthenticated."
+    )
+    chat_session_id: str = Field(
+        description="UUID of the chat session, formatted as a hyphenated lowercase string (e.g. '550e8400-e29b-41d4-a716-446655440000'). Always present — the session is guaranteed to exist by the time this hook fires."
+    )
+
+
+class QueryProcessingResponse(BaseModel):
+    # Intentionally permissive — customer endpoints may return extra fields.
+    query: str | None = Field(
+        default=None,
+        description=(
+            "The query to use in the pipeline. "
+            "Null, empty string, whitespace-only, or absent = reject the query."
+        ),
+    )
+    rejection_message: str | None = Field(
+        default=None,
+        description="Message shown to the user when the query is rejected. Falls back to a generic message if not provided.",
+    )
+
+
+class QueryProcessingSpec(HookPointSpec):
+    """Hook point that runs on every user query before it enters the pipeline.
+
+    Call site: inside handle_stream_message_objects() in
+    backend/onyx/chat/process_message.py, immediately after message_text is
+    assigned from the request and before create_new_chat_message() saves it.
+
+    This is the earliest possible point in the query pipeline:
+    - Raw query — unmodified, exactly as the user typed it
+    - No side effects yet — message has not been saved to DB
+    - User identity is available for user-specific logic
+
+    Supported use cases:
+    - Query rejection: block queries based on content or user context
+    - Query rewriting: normalize, expand, or modify the query
+    - PII removal: scrub sensitive data before the LLM sees it
+    - Access control: reject queries from certain users or groups
+    - Query auditing: log or track queries based on business rules
+    """
+
+    hook_point = HookPoint.QUERY_PROCESSING
+    display_name = "Query Processing"
+    description = (
+        "Runs on every user query before it enters the pipeline. "
+        "Allows rewriting, filtering, or rejecting queries."
+    )
+    default_timeout_seconds = 5.0  # user is actively waiting — keep tight
+    fail_hard_description = (
+        "The query will be blocked and the user will see an error message."
+    )
+    default_fail_strategy = HookFailStrategy.HARD
+
+    payload_model = QueryProcessingPayload
+    response_model = QueryProcessingResponse

backend/onyx/hooks/registry.py

@@ -0,0 +1,45 @@
+from onyx.db.enums import HookPoint
+from onyx.hooks.points.base import HookPointSpec
+from onyx.hooks.points.document_ingestion import DocumentIngestionSpec
+from onyx.hooks.points.query_processing import QueryProcessingSpec
+
+# Internal: use `monkeypatch.setattr(registry_module, "_REGISTRY", {...})` to override in tests.
+_REGISTRY: dict[HookPoint, HookPointSpec] = {
+    HookPoint.DOCUMENT_INGESTION: DocumentIngestionSpec(),
+    HookPoint.QUERY_PROCESSING: QueryProcessingSpec(),
+}
+
+
+def validate_registry() -> None:
+    """Assert that every HookPoint enum value has a registered spec.
+
+    Call once at application startup (e.g. from the FastAPI lifespan hook).
+    Raises RuntimeError if any hook point is missing a spec.
+    """
+    missing = set(HookPoint) - set(_REGISTRY)
+    if missing:
+        raise RuntimeError(
+            f"Hook point(s) have no registered spec: {missing}. "
+            "Add an entry to onyx.hooks.registry._REGISTRY."
+        )
+
+
+def get_hook_point_spec(hook_point: HookPoint) -> HookPointSpec:
+    """Returns the spec for a given hook point.
+
+    Raises ValueError if the hook point has no registered spec — this is a
+    programmer error; every HookPoint enum value must have a corresponding spec
+    in _REGISTRY.
+    """
+    try:
+        return _REGISTRY[hook_point]
+    except KeyError:
+        raise ValueError(
+            f"No spec registered for hook point {hook_point!r}. "
+            "Add an entry to onyx.hooks.registry._REGISTRY."
+        )
+
+
+def get_all_specs() -> list[HookPointSpec]:
+    """Returns the specs for all registered hook points."""
+    return list(_REGISTRY.values())

backend/onyx/hooks/utils.py

@@ -0,0 +1,5 @@
+from onyx.configs.app_configs import HOOK_ENABLED
+from shared_configs.configs import MULTI_TENANT
+
+# True only when hooks are available: single-tenant deployment with HOOK_ENABLED=true.
+HOOKS_AVAILABLE: bool = HOOK_ENABLED and not MULTI_TENANT

backend/onyx/indexing/indexing_pipeline.py

@@ -395,6 +395,12 @@ def process_image_sections(documents: list[Document]) -> list[IndexingDocument]:
         llm = get_default_llm_with_vision()
 
     if not llm:
+        if get_image_extraction_and_analysis_enabled():
+            logger.warning(
+                "Image analysis is enabled but no vision-capable LLM is "
+                "available — images will not be summarized. Configure a "
+                "vision model in the admin LLM settings."
+            )
         # Even without LLM, we still convert to IndexingDocument with base Sections
         return [
             IndexingDocument(

backend/onyx/llm/factory.py

@@ -168,10 +168,23 @@ def get_default_llm_with_vision(
             if model_supports_image_input(
                 default_model.name, default_model.llm_provider.provider
             ):
+                logger.info(
+                    "Using default vision model: %s (provider=%s)",
+                    default_model.name,
+                    default_model.llm_provider.provider,
+                )
                 return create_vision_llm(
                     LLMProviderView.from_model(default_model.llm_provider),
                     default_model.name,
                 )
+            else:
+                logger.warning(
+                    "Default vision model %s (provider=%s) does not support "
+                    "image input — falling back to searching all providers",
+                    default_model.name,
+                    default_model.llm_provider.provider,
+                )
+
         # Fall back to searching all providers
         models = fetch_existing_models(
             db_session=db_session,
@@ -179,6 +192,10 @@ def get_default_llm_with_vision(
         )
 
         if not models:
+            logger.warning(
+                "No LLM models with VISION or CHAT flow type found — "
+                "image summarization will be disabled"
+            )
             return None
 
         for model in models:
@@ -200,11 +217,25 @@ def get_default_llm_with_vision(
 
     for model in sorted_models:
         if model_supports_image_input(model.name, model.llm_provider.provider):
+            logger.info(
+                "Using fallback vision model: %s (provider=%s)",
+                model.name,
+                model.llm_provider.provider,
+            )
             return create_vision_llm(
                 provider_map[model.llm_provider_id],
                 model.name,
             )
 
+    checked_models = [
+        f"{m.name} (provider={m.llm_provider.provider})" for m in sorted_models
+    ]
+    logger.warning(
+        "No vision-capable model found among %d candidates: %s — "
+        "image summarization will be disabled",
+        len(sorted_models),
+        ", ".join(checked_models),
+    )
     return None
 
 

backend/onyx/llm/multi_llm.py

@@ -530,6 +530,11 @@ class LitellmLLM(LLM):
                 ):
                     messages = _strip_tool_content_from_messages(messages)
 
+                # Only pass tool_choice when tools are present — some providers (e.g. Fireworks)
+                # reject requests where tool_choice is explicitly null.
+                if tools and tool_choice is not None:
+                    optional_kwargs["tool_choice"] = tool_choice
+
                 response = litellm.completion(
                     mock_response=get_llm_mock_response() or MOCK_LLM_RESPONSE,
                     model=model,
@@ -538,7 +543,6 @@ class LitellmLLM(LLM):
                     custom_llm_provider=self._custom_llm_provider or None,
                     messages=messages,
                     tools=tools,
-                    tool_choice=tool_choice,
                     stream=stream,
                     temperature=temperature,
                     timeout=timeout_override or self._timeout,

backend/onyx/llm/utils.py

@@ -219,13 +219,26 @@ def litellm_exception_to_error_msg(
             "ratelimiterror"
         ):
             upstream_detail = upstream_detail.split(":", 1)[1].strip()
-        error_msg = (
-            f"{provider_name} rate limit: {upstream_detail}"
-            if upstream_detail
-            else f"{provider_name} rate limit exceeded: Please slow down your requests and try again later."
-        )
-        error_code = "RATE_LIMIT"
-        is_retryable = True
+        upstream_detail_lower = upstream_detail.lower()
+        if (
+            "insufficient_quota" in upstream_detail_lower
+            or "exceeded your current quota" in upstream_detail_lower
+        ):
+            error_msg = (
+                f"{provider_name} quota exceeded: {upstream_detail}"
+                if upstream_detail
+                else f"{provider_name} quota exceeded: Verify billing and quota for this API key."
+            )
+            error_code = "BUDGET_EXCEEDED"
+            is_retryable = False
+        else:
+            error_msg = (
+                f"{provider_name} rate limit: {upstream_detail}"
+                if upstream_detail
+                else f"{provider_name} rate limit exceeded: Please slow down your requests and try again later."
+            )
+            error_code = "RATE_LIMIT"
+            is_retryable = True
     elif isinstance(core_exception, ServiceUnavailableError):
         provider_name = (
             llm.config.model_provider

backend/onyx/main.py

@@ -62,6 +62,7 @@ from onyx.db.engine.sql_engine import get_session_with_current_tenant
 from onyx.db.engine.sql_engine import SqlEngine
 from onyx.error_handling.exceptions import register_onyx_exception_handlers
 from onyx.file_store.file_store import get_default_file_store
+from onyx.hooks.registry import validate_registry
 from onyx.server.api_key.api import router as api_key_router
 from onyx.server.auth_check import check_router_auth
 from onyx.server.documents.cc_pair import router as cc_pair_router
@@ -76,6 +77,7 @@ from onyx.server.features.default_assistant.api import (
 )
 from onyx.server.features.document_set.api import router as document_set_router
 from onyx.server.features.hierarchy.api import router as hierarchy_router
+from onyx.server.features.hooks.api import router as hook_router
 from onyx.server.features.input_prompt.api import (
     admin_router as admin_input_prompt_router,
 )
@@ -308,6 +310,7 @@ def validate_no_vector_db_settings() -> None:
 async def lifespan(app: FastAPI) -> AsyncGenerator[None, None]:  # noqa: ARG001
     validate_no_vector_db_settings()
     validate_cache_backend_settings()
+    validate_registry()
 
     # Set recursion limit
     if SYSTEM_RECURSION_LIMIT is not None:
@@ -451,6 +454,7 @@ def get_application(lifespan_override: Lifespan | None = None) -> FastAPI:
 
     register_onyx_exception_handlers(application)
 
+    include_router_with_global_prefix_prepended(application, hook_router)
     include_router_with_global_prefix_prepended(application, password_router)
     include_router_with_global_prefix_prepended(application, chat_router)
     include_router_with_global_prefix_prepended(application, query_router)

backend/onyx/server/documents/connector.py

@@ -479,7 +479,9 @@ def is_zip_file(file: UploadFile) -> bool:
 
 
 def upload_files(
-    files: list[UploadFile], file_origin: FileOrigin = FileOrigin.CONNECTOR
+    files: list[UploadFile],
+    file_origin: FileOrigin = FileOrigin.CONNECTOR,
+    unzip: bool = True,
 ) -> FileUploadResponse:
 
     # Skip directories and known macOS metadata entries
@@ -502,31 +504,46 @@ def upload_files(
                 if seen_zip:
                     raise HTTPException(status_code=400, detail=SEEN_ZIP_DETAIL)
                 seen_zip = True
+
+                # Validate the zip by opening it (catches corrupt/non-zip files)
                 with zipfile.ZipFile(file.file, "r") as zf:
-                    zip_metadata_file_id = save_zip_metadata_to_file_store(
-                        zf, file_store
-                    )
-                    for file_info in zf.namelist():
-                        if zf.getinfo(file_info).is_dir():
-                            continue
-
-                        if not should_process_file(file_info):
-                            continue
-
-                        sub_file_bytes = zf.read(file_info)
-
-                        mime_type, __ = mimetypes.guess_type(file_info)
-                        if mime_type is None:
-                            mime_type = "application/octet-stream"
-
-                        file_id = file_store.save_file(
-                            content=BytesIO(sub_file_bytes),
-                            display_name=os.path.basename(file_info),
-                            file_origin=file_origin,
-                            file_type=mime_type,
+                    if unzip:
+                        zip_metadata_file_id = save_zip_metadata_to_file_store(
+                            zf, file_store
                         )
-                        deduped_file_paths.append(file_id)
-                        deduped_file_names.append(os.path.basename(file_info))
+                        for file_info in zf.namelist():
+                            if zf.getinfo(file_info).is_dir():
+                                continue
+
+                            if not should_process_file(file_info):
+                                continue
+
+                            sub_file_bytes = zf.read(file_info)
+
+                            mime_type, __ = mimetypes.guess_type(file_info)
+                            if mime_type is None:
+                                mime_type = "application/octet-stream"
+
+                            file_id = file_store.save_file(
+                                content=BytesIO(sub_file_bytes),
+                                display_name=os.path.basename(file_info),
+                                file_origin=file_origin,
+                                file_type=mime_type,
+                            )
+                            deduped_file_paths.append(file_id)
+                            deduped_file_names.append(os.path.basename(file_info))
+                        continue
+
+                # Store the zip as-is (unzip=False)
+                file.file.seek(0)
+                file_id = file_store.save_file(
+                    content=file.file,
+                    display_name=file.filename,
+                    file_origin=file_origin,
+                    file_type=file.content_type or "application/zip",
+                )
+                deduped_file_paths.append(file_id)
+                deduped_file_names.append(file.filename)
                 continue
 
             # Since we can't render docx files in the UI,
@@ -613,9 +630,10 @@ def _fetch_and_check_file_connector_cc_pair_permissions(
 @router.post("/admin/connector/file/upload", tags=PUBLIC_API_TAGS)
 def upload_files_api(
     files: list[UploadFile],
+    unzip: bool = True,
     _: User = Depends(current_curator_or_admin_user),
 ) -> FileUploadResponse:
-    return upload_files(files, FileOrigin.OTHER)
+    return upload_files(files, FileOrigin.OTHER, unzip=unzip)
 
 
 @router.get("/admin/connector/{connector_id}/files", tags=PUBLIC_API_TAGS)

backend/onyx/server/documents/models.py

@@ -408,7 +408,7 @@ class FailedConnectorIndexingStatus(BaseModel):
     """Simplified version of ConnectorIndexingStatus for failed indexing attempts"""
 
     cc_pair_id: int
-    name: str | None
+    name: str
     error_msg: str | None
     is_deletable: bool
     connector_id: int
@@ -422,7 +422,7 @@ class ConnectorStatus(BaseModel):
     """
 
     cc_pair_id: int
-    name: str | None
+    name: str
     connector: ConnectorSnapshot
     credential: CredentialSnapshot
     access_type: AccessType
@@ -453,7 +453,7 @@ class DocsCountOperator(str, Enum):
 
 class ConnectorIndexingStatusLite(BaseModel):
     cc_pair_id: int
-    name: str | None
+    name: str
     source: DocumentSource
     access_type: AccessType
     cc_pair_status: ConnectorCredentialPairStatus
@@ -488,7 +488,7 @@ class ConnectorCredentialPairIdentifier(BaseModel):
 
 
 class ConnectorCredentialPairMetadata(BaseModel):
-    name: str | None = None
+    name: str
     access_type: AccessType
     auto_sync_options: dict[str, Any] | None = None
     groups: list[int] = Field(default_factory=list)
@@ -501,7 +501,7 @@ class CCStatusUpdateRequest(BaseModel):
 
 class ConnectorCredentialPairDescriptor(BaseModel):
     id: int
-    name: str | None = None
+    name: str
     connector: ConnectorSnapshot
     credential: CredentialSnapshot
     access_type: AccessType
@@ -511,7 +511,7 @@ class CCPairSummary(BaseModel):
     """Simplified connector-credential pair information with just essential data"""
 
     id: int
-    name: str | None
+    name: str
     source: DocumentSource
     access_type: AccessType
 

backend/onyx/server/features/build/sandbox/kubernetes/docker/templates/outputs/web/package-lock.json

@@ -15,7 +15,7 @@
         "date-fns": "^4.1.0",
         "embla-carousel-react": "^8.6.0",
         "lucide-react": "^0.562.0",
-        "next": "16.1.5",
+        "next": "16.1.7",
         "next-themes": "^0.4.6",
         "radix-ui": "^1.4.3",
         "react": "19.2.3",
@@ -1711,9 +1711,9 @@
       }
     },
     "node_modules/@next/env": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/env/-/env-16.1.5.tgz",
-      "integrity": "sha512-CRSCPJiSZoi4Pn69RYBDI9R7YK2g59vLexPQFXY0eyw+ILevIenCywzg+DqmlBik9zszEnw2HLFOUlLAcJbL7g==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/env/-/env-16.1.7.tgz",
+      "integrity": "sha512-rJJbIdJB/RQr2F1nylZr/PJzamvNNhfr3brdKP6s/GW850jbtR70QlSfFselvIBbcPUOlQwBakexjFzqLzF6pg==",
       "license": "MIT"
     },
     "node_modules/@next/eslint-plugin-next": {
@@ -1727,9 +1727,9 @@
       }
     },
     "node_modules/@next/swc-darwin-arm64": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/swc-darwin-arm64/-/swc-darwin-arm64-16.1.5.tgz",
-      "integrity": "sha512-eK7Wdm3Hjy/SCL7TevlH0C9chrpeOYWx2iR7guJDaz4zEQKWcS1IMVfMb9UKBFMg1XgzcPTYPIp1Vcpukkjg6Q==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/swc-darwin-arm64/-/swc-darwin-arm64-16.1.7.tgz",
+      "integrity": "sha512-b2wWIE8sABdyafc4IM8r5Y/dS6kD80JRtOGrUiKTsACFQfWWgUQ2NwoUX1yjFMXVsAwcQeNpnucF2ZrujsBBPg==",
       "cpu": [
         "arm64"
       ],
@@ -1743,9 +1743,9 @@
       }
     },
     "node_modules/@next/swc-darwin-x64": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/swc-darwin-x64/-/swc-darwin-x64-16.1.5.tgz",
-      "integrity": "sha512-foQscSHD1dCuxBmGkbIr6ScAUF6pRoDZP6czajyvmXPAOFNnQUJu2Os1SGELODjKp/ULa4fulnBWoHV3XdPLfA==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/swc-darwin-x64/-/swc-darwin-x64-16.1.7.tgz",
+      "integrity": "sha512-zcnVaaZulS1WL0Ss38R5Q6D2gz7MtBu8GZLPfK+73D/hp4GFMrC2sudLky1QibfV7h6RJBJs/gOFvYP0X7UVlQ==",
       "cpu": [
         "x64"
       ],
@@ -1759,9 +1759,9 @@
       }
     },
     "node_modules/@next/swc-linux-arm64-gnu": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/swc-linux-arm64-gnu/-/swc-linux-arm64-gnu-16.1.5.tgz",
-      "integrity": "sha512-qNIb42o3C02ccIeSeKjacF3HXotGsxh/FMk/rSRmCzOVMtoWH88odn2uZqF8RLsSUWHcAqTgYmPD3pZ03L9ZAA==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/swc-linux-arm64-gnu/-/swc-linux-arm64-gnu-16.1.7.tgz",
+      "integrity": "sha512-2ant89Lux/Q3VyC8vNVg7uBaFVP9SwoK2jJOOR0L8TQnX8CAYnh4uctAScy2Hwj2dgjVHqHLORQZJ2wH6VxhSQ==",
       "cpu": [
         "arm64"
       ],
@@ -1775,9 +1775,9 @@
       }
     },
     "node_modules/@next/swc-linux-arm64-musl": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/swc-linux-arm64-musl/-/swc-linux-arm64-musl-16.1.5.tgz",
-      "integrity": "sha512-U+kBxGUY1xMAzDTXmuVMfhaWUZQAwzRaHJ/I6ihtR5SbTVUEaDRiEU9YMjy1obBWpdOBuk1bcm+tsmifYSygfw==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/swc-linux-arm64-musl/-/swc-linux-arm64-musl-16.1.7.tgz",
+      "integrity": "sha512-uufcze7LYv0FQg9GnNeZ3/whYfo+1Q3HnQpm16o6Uyi0OVzLlk2ZWoY7j07KADZFY8qwDbsmFnMQP3p3+Ftprw==",
       "cpu": [
         "arm64"
       ],
@@ -1791,9 +1791,9 @@
       }
     },
     "node_modules/@next/swc-linux-x64-gnu": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/swc-linux-x64-gnu/-/swc-linux-x64-gnu-16.1.5.tgz",
-      "integrity": "sha512-gq2UtoCpN7Ke/7tKaU7i/1L7eFLfhMbXjNghSv0MVGF1dmuoaPeEVDvkDuO/9LVa44h5gqpWeJ4mRRznjDv7LA==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/swc-linux-x64-gnu/-/swc-linux-x64-gnu-16.1.7.tgz",
+      "integrity": "sha512-KWVf2gxYvHtvuT+c4MBOGxuse5TD7DsMFYSxVxRBnOzok/xryNeQSjXgxSv9QpIVlaGzEn/pIuI6Koosx8CGWA==",
       "cpu": [
         "x64"
       ],
@@ -1807,9 +1807,9 @@
       }
     },
     "node_modules/@next/swc-linux-x64-musl": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/swc-linux-x64-musl/-/swc-linux-x64-musl-16.1.5.tgz",
-      "integrity": "sha512-bQWSE729PbXT6mMklWLf8dotislPle2L70E9q6iwETYEOt092GDn0c+TTNj26AjmeceSsC4ndyGsK5nKqHYXjQ==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/swc-linux-x64-musl/-/swc-linux-x64-musl-16.1.7.tgz",
+      "integrity": "sha512-HguhaGwsGr1YAGs68uRKc4aGWxLET+NevJskOcCAwXbwj0fYX0RgZW2gsOCzr9S11CSQPIkxmoSbuVaBp4Z3dA==",
       "cpu": [
         "x64"
       ],
@@ -1823,9 +1823,9 @@
       }
     },
     "node_modules/@next/swc-win32-arm64-msvc": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/swc-win32-arm64-msvc/-/swc-win32-arm64-msvc-16.1.5.tgz",
-      "integrity": "sha512-LZli0anutkIllMtTAWZlDqdfvjWX/ch8AFK5WgkNTvaqwlouiD1oHM+WW8RXMiL0+vAkAJyAGEzPPjO+hnrSNQ==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/swc-win32-arm64-msvc/-/swc-win32-arm64-msvc-16.1.7.tgz",
+      "integrity": "sha512-S0n3KrDJokKTeFyM/vGGGR8+pCmXYrjNTk2ZozOL1C/JFdfUIL9O1ATaJOl5r2POe56iRChbsszrjMAdWSv7kQ==",
       "cpu": [
         "arm64"
       ],
@@ -1839,9 +1839,9 @@
       }
     },
     "node_modules/@next/swc-win32-x64-msvc": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/@next/swc-win32-x64-msvc/-/swc-win32-x64-msvc-16.1.5.tgz",
-      "integrity": "sha512-7is37HJTNQGhjPpQbkKjKEboHYQnCgpVt/4rBrrln0D9nderNxZ8ZWs8w1fAtzUx7wEyYjQ+/13myFgFj6K2Ng==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/@next/swc-win32-x64-msvc/-/swc-win32-x64-msvc-16.1.7.tgz",
+      "integrity": "sha512-mwgtg8CNZGYm06LeEd+bNnOUfwOyNem/rOiP14Lsz+AnUY92Zq/LXwtebtUiaeVkhbroRCQ0c8GlR4UT1U+0yg==",
       "cpu": [
         "x64"
       ],
@@ -4971,12 +4971,15 @@
       "license": "MIT"
     },
     "node_modules/baseline-browser-mapping": {
-      "version": "2.9.17",
-      "resolved": "https://registry.npmjs.org/baseline-browser-mapping/-/baseline-browser-mapping-2.9.17.tgz",
-      "integrity": "sha512-agD0MgJFUP/4nvjqzIB29zRPUuCF7Ge6mEv9s8dHrtYD7QWXRcx75rOADE/d5ah1NI+0vkDl0yorDd5U852IQQ==",
+      "version": "2.10.8",
+      "resolved": "https://registry.npmjs.org/baseline-browser-mapping/-/baseline-browser-mapping-2.10.8.tgz",
+      "integrity": "sha512-PCLz/LXGBsNTErbtB6i5u4eLpHeMfi93aUv5duMmj6caNu6IphS4q6UevDnL36sZQv9lrP11dbPKGMaXPwMKfQ==",
       "license": "Apache-2.0",
       "bin": {
-        "baseline-browser-mapping": "dist/cli.js"
+        "baseline-browser-mapping": "dist/cli.cjs"
+      },
+      "engines": {
+        "node": ">=6.0.0"
       }
     },
     "node_modules/body-parser": {
@@ -8975,14 +8978,14 @@
       }
     },
     "node_modules/next": {
-      "version": "16.1.5",
-      "resolved": "https://registry.npmjs.org/next/-/next-16.1.5.tgz",
-      "integrity": "sha512-f+wE+NSbiQgh3DSAlTaw2FwY5yGdVViAtp8TotNQj4kk4Q8Bh1sC/aL9aH+Rg1YAVn18OYXsRDT7U/079jgP7w==",
+      "version": "16.1.7",
+      "resolved": "https://registry.npmjs.org/next/-/next-16.1.7.tgz",
+      "integrity": "sha512-WM0L7WrSvKwoLegLYr6V+mz+RIofqQgVAfHhMp9a88ms0cFX8iX9ew+snpWlSBwpkURJOUdvCEt3uLl3NNzvWg==",
       "license": "MIT",
       "dependencies": {
-        "@next/env": "16.1.5",
+        "@next/env": "16.1.7",
         "@swc/helpers": "0.5.15",
-        "baseline-browser-mapping": "^2.8.3",
+        "baseline-browser-mapping": "^2.9.19",
         "caniuse-lite": "^1.0.30001579",
         "postcss": "8.4.31",
         "styled-jsx": "5.1.6"
@@ -8994,14 +8997,14 @@
         "node": ">=20.9.0"
       },
       "optionalDependencies": {
-        "@next/swc-darwin-arm64": "16.1.5",
-        "@next/swc-darwin-x64": "16.1.5",
-        "@next/swc-linux-arm64-gnu": "16.1.5",
-        "@next/swc-linux-arm64-musl": "16.1.5",
-        "@next/swc-linux-x64-gnu": "16.1.5",
-        "@next/swc-linux-x64-musl": "16.1.5",
-        "@next/swc-win32-arm64-msvc": "16.1.5",
-        "@next/swc-win32-x64-msvc": "16.1.5",
+        "@next/swc-darwin-arm64": "16.1.7",
+        "@next/swc-darwin-x64": "16.1.7",
+        "@next/swc-linux-arm64-gnu": "16.1.7",
+        "@next/swc-linux-arm64-musl": "16.1.7",
+        "@next/swc-linux-x64-gnu": "16.1.7",
+        "@next/swc-linux-x64-musl": "16.1.7",
+        "@next/swc-win32-arm64-msvc": "16.1.7",
+        "@next/swc-win32-x64-msvc": "16.1.7",
         "sharp": "^0.34.4"
       },
       "peerDependencies": {

backend/onyx/server/features/build/sandbox/kubernetes/docker/templates/outputs/web/package.json

@@ -16,7 +16,7 @@
     "date-fns": "^4.1.0",
     "embla-carousel-react": "^8.6.0",
     "lucide-react": "^0.562.0",
-    "next": "16.1.5",
+    "next": "16.1.7",
     "next-themes": "^0.4.6",
     "radix-ui": "^1.4.3",
     "react": "19.2.3",

backend/onyx/server/features/hooks/api.py

@@ -0,0 +1,453 @@
+import httpx
+from fastapi import APIRouter
+from fastapi import Depends
+from fastapi import Query
+from sqlalchemy.orm import Session
+
+from onyx.auth.users import current_admin_user
+from onyx.auth.users import User
+from onyx.db.constants import UNSET
+from onyx.db.constants import UnsetType
+from onyx.db.engine.sql_engine import get_session
+from onyx.db.engine.sql_engine import get_session_with_current_tenant
+from onyx.db.hook import create_hook__no_commit
+from onyx.db.hook import delete_hook__no_commit
+from onyx.db.hook import get_hook_by_id
+from onyx.db.hook import get_hook_execution_logs
+from onyx.db.hook import get_hooks
+from onyx.db.hook import update_hook__no_commit
+from onyx.db.models import Hook
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
+from onyx.hooks.api_dependencies import require_hook_enabled
+from onyx.hooks.models import HookCreateRequest
+from onyx.hooks.models import HookExecutionRecord
+from onyx.hooks.models import HookPointMetaResponse
+from onyx.hooks.models import HookResponse
+from onyx.hooks.models import HookUpdateRequest
+from onyx.hooks.models import HookValidateResponse
+from onyx.hooks.models import HookValidateStatus
+from onyx.hooks.registry import get_all_specs
+from onyx.hooks.registry import get_hook_point_spec
+from onyx.utils.logger import setup_logger
+from onyx.utils.url import SSRFException
+from onyx.utils.url import validate_outbound_http_url
+
+logger = setup_logger()
+
+# ---------------------------------------------------------------------------
+# SSRF protection
+# ---------------------------------------------------------------------------
+
+
+def _check_ssrf_safety(endpoint_url: str) -> None:
+    """Raise OnyxError if endpoint_url could be used for SSRF.
+
+    Delegates to validate_outbound_http_url with https_only=True.
+    """
+    try:
+        validate_outbound_http_url(endpoint_url, https_only=True)
+    except (SSRFException, ValueError) as e:
+        raise OnyxError(OnyxErrorCode.INVALID_INPUT, str(e))
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+
+def _hook_to_response(hook: Hook, creator_email: str | None = None) -> HookResponse:
+    return HookResponse(
+        id=hook.id,
+        name=hook.name,
+        hook_point=hook.hook_point,
+        endpoint_url=hook.endpoint_url,
+        fail_strategy=hook.fail_strategy,
+        timeout_seconds=hook.timeout_seconds,
+        is_active=hook.is_active,
+        is_reachable=hook.is_reachable,
+        creator_email=(
+            creator_email
+            if creator_email is not None
+            else (hook.creator.email if hook.creator else None)
+        ),
+        created_at=hook.created_at,
+        updated_at=hook.updated_at,
+    )
+
+
+def _get_hook_or_404(
+    db_session: Session,
+    hook_id: int,
+    include_creator: bool = False,
+) -> Hook:
+    hook = get_hook_by_id(
+        db_session=db_session,
+        hook_id=hook_id,
+        include_creator=include_creator,
+    )
+    if hook is None:
+        raise OnyxError(OnyxErrorCode.NOT_FOUND, f"Hook {hook_id} not found.")
+    return hook
+
+
+def _raise_for_validation_failure(validation: HookValidateResponse) -> None:
+    """Raise an appropriate OnyxError for a non-passed validation result."""
+    if validation.status == HookValidateStatus.auth_failed:
+        raise OnyxError(OnyxErrorCode.CREDENTIAL_INVALID, validation.error_message)
+    if validation.status == HookValidateStatus.timeout:
+        raise OnyxError(
+            OnyxErrorCode.GATEWAY_TIMEOUT,
+            f"Endpoint validation failed: {validation.error_message}",
+        )
+    raise OnyxError(
+        OnyxErrorCode.BAD_GATEWAY,
+        f"Endpoint validation failed: {validation.error_message}",
+    )
+
+
+def _validate_endpoint(
+    endpoint_url: str,
+    api_key: str | None,
+    timeout_seconds: float,
+) -> HookValidateResponse:
+    """Check whether endpoint_url is reachable by sending an empty POST request.
+
+    We use POST since hook endpoints expect POST requests. The server will typically
+    respond with 4xx (missing/invalid body) — that is fine. Any HTTP response means
+    the server is up and routable. A 401/403 response returns auth_failed
+    (not reachable — indicates the api_key is invalid).
+
+    Timeout handling:
+    - ConnectTimeout: TCP handshake never completed → cannot_connect.
+    - ReadTimeout / WriteTimeout: TCP was established, server responded slowly → timeout
+      (operator should consider increasing timeout_seconds).
+    - All other exceptions → cannot_connect.
+    """
+    _check_ssrf_safety(endpoint_url)
+    headers: dict[str, str] = {}
+    if api_key:
+        headers["Authorization"] = f"Bearer {api_key}"
+    try:
+        with httpx.Client(timeout=timeout_seconds, follow_redirects=False) as client:
+            response = client.post(endpoint_url, headers=headers)
+        if response.status_code in (401, 403):
+            return HookValidateResponse(
+                status=HookValidateStatus.auth_failed,
+                error_message=f"Authentication failed (HTTP {response.status_code})",
+            )
+        return HookValidateResponse(status=HookValidateStatus.passed)
+    except httpx.TimeoutException as exc:
+        # ConnectTimeout: TCP handshake never completed → cannot_connect.
+        # ReadTimeout / WriteTimeout: TCP was established, server just responded slowly → timeout.
+        if isinstance(exc, httpx.ConnectTimeout):
+            logger.warning(
+                "Hook endpoint validation: connect timeout for %s",
+                endpoint_url,
+                exc_info=exc,
+            )
+            return HookValidateResponse(
+                status=HookValidateStatus.cannot_connect, error_message=str(exc)
+            )
+        logger.warning(
+            "Hook endpoint validation: read/write timeout for %s",
+            endpoint_url,
+            exc_info=exc,
+        )
+        return HookValidateResponse(
+            status=HookValidateStatus.timeout,
+            error_message="Endpoint timed out — consider increasing timeout_seconds.",
+        )
+    except Exception as exc:
+        logger.warning(
+            "Hook endpoint validation: connection error for %s",
+            endpoint_url,
+            exc_info=exc,
+        )
+        return HookValidateResponse(
+            status=HookValidateStatus.cannot_connect, error_message=str(exc)
+        )
+
+
+# ---------------------------------------------------------------------------
+# Routers
+# ---------------------------------------------------------------------------
+
+router = APIRouter(prefix="/admin/hooks")
+
+
+# ---------------------------------------------------------------------------
+# Hook endpoints
+# ---------------------------------------------------------------------------
+
+
+@router.get("/specs")
+def get_hook_point_specs(
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+) -> list[HookPointMetaResponse]:
+    return [
+        HookPointMetaResponse(
+            hook_point=spec.hook_point,
+            display_name=spec.display_name,
+            description=spec.description,
+            docs_url=spec.docs_url,
+            input_schema=spec.input_schema,
+            output_schema=spec.output_schema,
+            default_timeout_seconds=spec.default_timeout_seconds,
+            default_fail_strategy=spec.default_fail_strategy,
+            fail_hard_description=spec.fail_hard_description,
+        )
+        for spec in get_all_specs()
+    ]
+
+
+@router.get("")
+def list_hooks(
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> list[HookResponse]:
+    hooks = get_hooks(db_session=db_session, include_creator=True)
+    return [_hook_to_response(h) for h in hooks]
+
+
+@router.post("")
+def create_hook(
+    req: HookCreateRequest,
+    user: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> HookResponse:
+    """Create a new hook. The endpoint is validated before persisting — creation fails if
+    the endpoint cannot be reached or the api_key is invalid. Hooks are created inactive;
+    use POST /{hook_id}/activate once ready to receive traffic."""
+    spec = get_hook_point_spec(req.hook_point)
+    api_key = req.api_key.get_secret_value() if req.api_key else None
+    validation = _validate_endpoint(
+        endpoint_url=req.endpoint_url,
+        api_key=api_key,
+        timeout_seconds=req.timeout_seconds or spec.default_timeout_seconds,
+    )
+    if validation.status != HookValidateStatus.passed:
+        _raise_for_validation_failure(validation)
+
+    hook = create_hook__no_commit(
+        db_session=db_session,
+        name=req.name,
+        hook_point=req.hook_point,
+        endpoint_url=req.endpoint_url,
+        api_key=api_key,
+        fail_strategy=req.fail_strategy or spec.default_fail_strategy,
+        timeout_seconds=req.timeout_seconds or spec.default_timeout_seconds,
+        creator_id=user.id,
+    )
+    hook.is_reachable = True
+    db_session.commit()
+    return _hook_to_response(hook, creator_email=user.email)
+
+
+@router.get("/{hook_id}")
+def get_hook(
+    hook_id: int,
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> HookResponse:
+    hook = _get_hook_or_404(db_session, hook_id, include_creator=True)
+    return _hook_to_response(hook)
+
+
+@router.patch("/{hook_id}")
+def update_hook(
+    hook_id: int,
+    req: HookUpdateRequest,
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> HookResponse:
+    """Update hook fields. If endpoint_url, api_key, or timeout_seconds changes, the
+    endpoint is re-validated using the effective values. For active hooks the update is
+    rejected on validation failure, keeping live traffic unaffected. For inactive hooks
+    the update goes through regardless and is_reachable is updated to reflect the result.
+
+    Note: if an active hook's endpoint is currently down, even a timeout_seconds-only
+    increase will be rejected. The recovery flow is: deactivate → update → reactivate.
+    """
+    # api_key: UNSET = no change, None = clear, value = update
+    api_key: str | None | UnsetType
+    if "api_key" not in req.model_fields_set:
+        api_key = UNSET
+    elif req.api_key is None:
+        api_key = None
+    else:
+        api_key = req.api_key.get_secret_value()
+
+    endpoint_url_changing = "endpoint_url" in req.model_fields_set
+    api_key_changing = not isinstance(api_key, UnsetType)
+    timeout_changing = "timeout_seconds" in req.model_fields_set
+
+    validated_is_reachable: bool | None = None
+    if endpoint_url_changing or api_key_changing or timeout_changing:
+        existing = _get_hook_or_404(db_session, hook_id)
+        effective_url: str = (
+            req.endpoint_url if endpoint_url_changing else existing.endpoint_url  # type: ignore[assignment]  # endpoint_url is required on create and cannot be cleared on update
+        )
+        effective_api_key: str | None = (
+            (api_key if not isinstance(api_key, UnsetType) else None)
+            if api_key_changing
+            else (
+                existing.api_key.get_value(apply_mask=False)
+                if existing.api_key
+                else None
+            )
+        )
+        effective_timeout: float = (
+            req.timeout_seconds if timeout_changing else existing.timeout_seconds  # type: ignore[assignment]  # req.timeout_seconds is non-None when timeout_changing (validated by HookUpdateRequest)
+        )
+        validation = _validate_endpoint(
+            endpoint_url=effective_url,
+            api_key=effective_api_key,
+            timeout_seconds=effective_timeout,
+        )
+        if existing.is_active and validation.status != HookValidateStatus.passed:
+            _raise_for_validation_failure(validation)
+        validated_is_reachable = validation.status == HookValidateStatus.passed
+
+    hook = update_hook__no_commit(
+        db_session=db_session,
+        hook_id=hook_id,
+        name=req.name,
+        endpoint_url=(req.endpoint_url if endpoint_url_changing else UNSET),
+        api_key=api_key,
+        fail_strategy=req.fail_strategy,
+        timeout_seconds=req.timeout_seconds,
+        is_reachable=validated_is_reachable,
+        include_creator=True,
+    )
+    db_session.commit()
+    return _hook_to_response(hook)
+
+
+@router.delete("/{hook_id}")
+def delete_hook(
+    hook_id: int,
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> None:
+    delete_hook__no_commit(db_session=db_session, hook_id=hook_id)
+    db_session.commit()
+
+
+@router.post("/{hook_id}/activate")
+def activate_hook(
+    hook_id: int,
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> HookResponse:
+    hook = _get_hook_or_404(db_session, hook_id)
+    if not hook.endpoint_url:
+        raise OnyxError(
+            OnyxErrorCode.INVALID_INPUT, "Hook has no endpoint URL configured."
+        )
+
+    api_key = hook.api_key.get_value(apply_mask=False) if hook.api_key else None
+    validation = _validate_endpoint(
+        endpoint_url=hook.endpoint_url,
+        api_key=api_key,
+        timeout_seconds=hook.timeout_seconds,
+    )
+    if validation.status != HookValidateStatus.passed:
+        # Persist is_reachable=False in a separate session so the request
+        # session has no commits on the failure path and the transaction
+        # boundary stays clean.
+        if hook.is_reachable is not False:
+            with get_session_with_current_tenant() as side_session:
+                update_hook__no_commit(
+                    db_session=side_session, hook_id=hook_id, is_reachable=False
+                )
+                side_session.commit()
+        _raise_for_validation_failure(validation)
+
+    hook = update_hook__no_commit(
+        db_session=db_session,
+        hook_id=hook_id,
+        is_active=True,
+        is_reachable=True,
+        include_creator=True,
+    )
+    db_session.commit()
+    return _hook_to_response(hook)
+
+
+@router.post("/{hook_id}/validate")
+def validate_hook(
+    hook_id: int,
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> HookValidateResponse:
+    hook = _get_hook_or_404(db_session, hook_id)
+    if not hook.endpoint_url:
+        raise OnyxError(
+            OnyxErrorCode.INVALID_INPUT, "Hook has no endpoint URL configured."
+        )
+
+    api_key = hook.api_key.get_value(apply_mask=False) if hook.api_key else None
+    validation = _validate_endpoint(
+        endpoint_url=hook.endpoint_url,
+        api_key=api_key,
+        timeout_seconds=hook.timeout_seconds,
+    )
+    validation_passed = validation.status == HookValidateStatus.passed
+    if hook.is_reachable != validation_passed:
+        update_hook__no_commit(
+            db_session=db_session, hook_id=hook_id, is_reachable=validation_passed
+        )
+        db_session.commit()
+    return validation
+
+
+@router.post("/{hook_id}/deactivate")
+def deactivate_hook(
+    hook_id: int,
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> HookResponse:
+    hook = update_hook__no_commit(
+        db_session=db_session,
+        hook_id=hook_id,
+        is_active=False,
+        include_creator=True,
+    )
+    db_session.commit()
+    return _hook_to_response(hook)
+
+
+# ---------------------------------------------------------------------------
+# Execution log endpoints
+# ---------------------------------------------------------------------------
+
+
+@router.get("/{hook_id}/execution-logs")
+def list_hook_execution_logs(
+    hook_id: int,
+    limit: int = Query(default=10, ge=1, le=100),
+    _: User = Depends(current_admin_user),
+    _hook_enabled: None = Depends(require_hook_enabled),
+    db_session: Session = Depends(get_session),
+) -> list[HookExecutionRecord]:
+    _get_hook_or_404(db_session, hook_id)
+    logs = get_hook_execution_logs(db_session=db_session, hook_id=hook_id, limit=limit)
+    return [
+        HookExecutionRecord(
+            error_message=log.error_message,
+            status_code=log.status_code,
+            duration_ms=log.duration_ms,
+            created_at=log.created_at,
+        )
+        for log in logs
+    ]

backend/onyx/server/query_and_chat/streaming_utils.py

@@ -1,239 +0,0 @@
-from __future__ import annotations
-
-import re
-
-from onyx.context.search.models import SavedSearchDoc
-from onyx.context.search.models import SearchDoc
-from onyx.server.query_and_chat.placement import Placement
-from onyx.server.query_and_chat.streaming_models import AgentResponseDelta
-from onyx.server.query_and_chat.streaming_models import AgentResponseStart
-from onyx.server.query_and_chat.streaming_models import CitationInfo
-from onyx.server.query_and_chat.streaming_models import GeneratedImage
-from onyx.server.query_and_chat.streaming_models import ImageGenerationFinal
-from onyx.server.query_and_chat.streaming_models import ImageGenerationToolStart
-from onyx.server.query_and_chat.streaming_models import OpenUrlDocuments
-from onyx.server.query_and_chat.streaming_models import OpenUrlStart
-from onyx.server.query_and_chat.streaming_models import OpenUrlUrls
-from onyx.server.query_and_chat.streaming_models import Packet
-from onyx.server.query_and_chat.streaming_models import ReasoningDelta
-from onyx.server.query_and_chat.streaming_models import ReasoningStart
-from onyx.server.query_and_chat.streaming_models import SearchToolDocumentsDelta
-from onyx.server.query_and_chat.streaming_models import SearchToolQueriesDelta
-from onyx.server.query_and_chat.streaming_models import SearchToolStart
-from onyx.server.query_and_chat.streaming_models import SectionEnd
-
-
-_CANNOT_SHOW_STEP_RESULTS_STR = "[Cannot display step results]"
-
-
-def _adjust_message_text_for_agent_search_results(
-    adjusted_message_text: str,
-    final_documents: list[SavedSearchDoc],  # noqa: ARG001
-) -> str:
-    # Remove all [Q<integer>] patterns (sub-question citations)
-    return re.sub(r"\[Q\d+\]", "", adjusted_message_text)
-
-
-def _replace_d_citations_with_links(
-    message_text: str, final_documents: list[SavedSearchDoc]
-) -> str:
-    def replace_citation(match: re.Match[str]) -> str:
-        d_number = match.group(1)
-        try:
-            doc_index = int(d_number) - 1
-            if 0 <= doc_index < len(final_documents):
-                doc = final_documents[doc_index]
-                link = doc.link if doc.link else ""
-                return f"[[{d_number}]]({link})"
-            return match.group(0)
-        except (ValueError, IndexError):
-            return match.group(0)
-
-    return re.sub(r"\[D(\d+)\]", replace_citation, message_text)
-
-
-def create_message_packets(
-    message_text: str,
-    final_documents: list[SavedSearchDoc] | None,
-    turn_index: int,
-    is_legacy_agentic: bool = False,
-) -> list[Packet]:
-    packets: list[Packet] = []
-
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=AgentResponseStart(
-                final_documents=SearchDoc.from_saved_search_docs(final_documents or []),
-            ),
-        )
-    )
-
-    adjusted_message_text = message_text
-    if is_legacy_agentic:
-        if final_documents is not None:
-            adjusted_message_text = _adjust_message_text_for_agent_search_results(
-                message_text, final_documents
-            )
-            adjusted_message_text = _replace_d_citations_with_links(
-                adjusted_message_text, final_documents
-            )
-        else:
-            adjusted_message_text = re.sub(r"\[Q\d+\]", "", message_text)
-
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=AgentResponseDelta(
-                content=adjusted_message_text,
-            ),
-        ),
-    )
-
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=SectionEnd(),
-        )
-    )
-
-    return packets
-
-
-def create_citation_packets(
-    citation_info_list: list[CitationInfo], turn_index: int
-) -> list[Packet]:
-    packets: list[Packet] = []
-
-    # Emit each citation as a separate CitationInfo packet
-    for citation_info in citation_info_list:
-        packets.append(
-            Packet(
-                placement=Placement(turn_index=turn_index),
-                obj=citation_info,
-            )
-        )
-
-    packets.append(Packet(placement=Placement(turn_index=turn_index), obj=SectionEnd()))
-
-    return packets
-
-
-def create_reasoning_packets(reasoning_text: str, turn_index: int) -> list[Packet]:
-    packets: list[Packet] = []
-
-    packets.append(
-        Packet(placement=Placement(turn_index=turn_index), obj=ReasoningStart())
-    )
-
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=ReasoningDelta(
-                reasoning=reasoning_text,
-            ),
-        ),
-    )
-
-    packets.append(Packet(placement=Placement(turn_index=turn_index), obj=SectionEnd()))
-
-    return packets
-
-
-def create_image_generation_packets(
-    images: list[GeneratedImage], turn_index: int
-) -> list[Packet]:
-    packets: list[Packet] = []
-
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=ImageGenerationToolStart(),
-        )
-    )
-
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=ImageGenerationFinal(images=images),
-        ),
-    )
-
-    packets.append(Packet(placement=Placement(turn_index=turn_index), obj=SectionEnd()))
-
-    return packets
-
-
-def create_fetch_packets(
-    fetch_docs: list[SavedSearchDoc],
-    urls: list[str],
-    turn_index: int,
-) -> list[Packet]:
-    packets: list[Packet] = []
-    # Emit start packet
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=OpenUrlStart(),
-        )
-    )
-    # Emit URLs packet
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=OpenUrlUrls(urls=urls),
-        )
-    )
-    # Emit documents packet
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=OpenUrlDocuments(
-                documents=SearchDoc.from_saved_search_docs(fetch_docs)
-            ),
-        )
-    )
-    packets.append(Packet(placement=Placement(turn_index=turn_index), obj=SectionEnd()))
-    return packets
-
-
-def create_search_packets(
-    search_queries: list[str],
-    saved_search_docs: list[SavedSearchDoc],
-    is_internet_search: bool,
-    turn_index: int,
-) -> list[Packet]:
-    packets: list[Packet] = []
-
-    packets.append(
-        Packet(
-            placement=Placement(turn_index=turn_index),
-            obj=SearchToolStart(
-                is_internet_search=is_internet_search,
-            ),
-        )
-    )
-
-    # Emit queries if present
-    if search_queries:
-        packets.append(
-            Packet(
-                placement=Placement(turn_index=turn_index),
-                obj=SearchToolQueriesDelta(queries=search_queries),
-            ),
-        )
-
-    # Emit documents if present
-    if saved_search_docs:
-        packets.append(
-            Packet(
-                placement=Placement(turn_index=turn_index),
-                obj=SearchToolDocumentsDelta(
-                    documents=SearchDoc.from_saved_search_docs(saved_search_docs)
-                ),
-            ),
-        )
-
-    packets.append(Packet(placement=Placement(turn_index=turn_index), obj=SectionEnd()))
-
-    return packets

backend/onyx/tools/tool_implementations/python/python_tool.py

@@ -1,3 +1,4 @@
+import hashlib
 import mimetypes
 from io import BytesIO
 from typing import Any
@@ -83,6 +84,14 @@ class PythonTool(Tool[PythonToolOverrideKwargs]):
     def __init__(self, tool_id: int, emitter: Emitter) -> None:
         super().__init__(emitter=emitter)
         self._id = tool_id
+        # Cache of (filename, content_hash) -> ci_file_id to avoid re-uploading
+        # the same file on every tool call iteration within the same agent session.
+        # Filename is included in the key so two files with identical bytes but
+        # different names each get their own upload slot.
+        # TTL assumption: code-interpreter file TTLs (typically hours) greatly
+        # exceed the lifetime of a single agent session (at most MAX_LLM_CYCLES
+        # iterations, typically a few minutes), so stale-ID eviction is not needed.
+        self._uploaded_file_cache: dict[tuple[str, str], str] = {}
 
     @property
     def id(self) -> int:
@@ -182,8 +191,13 @@ class PythonTool(Tool[PythonToolOverrideKwargs]):
             for ind, chat_file in enumerate(chat_files):
                 file_name = chat_file.filename or f"file_{ind}"
                 try:
-                    # Upload to Code Interpreter
-                    ci_file_id = client.upload_file(chat_file.content, file_name)
+                    content_hash = hashlib.sha256(chat_file.content).hexdigest()
+                    cache_key = (file_name, content_hash)
+                    ci_file_id = self._uploaded_file_cache.get(cache_key)
+                    if ci_file_id is None:
+                        # Upload to Code Interpreter
+                        ci_file_id = client.upload_file(chat_file.content, file_name)
+                        self._uploaded_file_cache[cache_key] = ci_file_id
 
                     # Stage for execution
                     files_to_stage.append({"path": file_name, "file_id": ci_file_id})
@@ -299,14 +313,10 @@ class PythonTool(Tool[PythonToolOverrideKwargs]):
                             f"Failed to delete Code Interpreter generated file {ci_file_id}: {e}"
                         )
 
-                # Cleanup staged input files
-                for file_mapping in files_to_stage:
-                    try:
-                        client.delete_file(file_mapping["file_id"])
-                    except Exception as e:
-                        logger.error(
-                            f"Failed to delete Code Interpreter staged file {file_mapping['file_id']}: {e}"
-                        )
+                # Note: staged input files are intentionally not deleted here because
+                # _uploaded_file_cache reuses their file_ids across iterations. They are
+                # orphaned when the session ends, but the code interpreter cleans up
+                # stale files on its own TTL.
 
                 # Emit file_ids once files are processed
                 if generated_file_ids:

backend/onyx/utils/middleware.py

@@ -74,7 +74,7 @@ def make_structured_onyx_request_id(prefix: str, request_url: str) -> str:
 
 def _make_onyx_request_id(prefix: str, hash_input: str) -> str:
     """helper function to return an id given a string input"""
-    hash_obj = hashlib.md5(hash_input.encode("utf-8"))
+    hash_obj = hashlib.md5(hash_input.encode("utf-8"), usedforsecurity=False)
     hash_bytes = hash_obj.digest()[:6]  # Truncate to 6 bytes
 
     # 6 bytes becomes 8 bytes. we shouldn't need to strip but just in case

backend/onyx/utils/url.py

@@ -140,10 +140,20 @@ def _validate_and_resolve_url(url: str) -> tuple[str, str, int]:
     return validated_ip, hostname, port
 
 
-def validate_outbound_http_url(url: str, *, allow_private_network: bool = False) -> str:
+def validate_outbound_http_url(
+    url: str,
+    *,
+    allow_private_network: bool = False,
+    https_only: bool = False,
+) -> str:
     """
     Validate a URL that will be used by backend outbound HTTP calls.
 
+    Args:
+        url: The URL to validate.
+        allow_private_network: If True, skip private/reserved IP checks.
+        https_only: If True, reject http:// URLs (only https:// is allowed).
+
     Returns:
         A normalized URL string with surrounding whitespace removed.
 
@@ -157,7 +167,12 @@ def validate_outbound_http_url(url: str, *, allow_private_network: bool = False)
 
     parsed = urlparse(normalized_url)
 
-    if parsed.scheme not in ("http", "https"):
+    if https_only:
+        if parsed.scheme != "https":
+            raise SSRFException(
+                f"Invalid URL scheme '{parsed.scheme}'. Only https is allowed."
+            )
+    elif parsed.scheme not in ("http", "https"):
         raise SSRFException(
             f"Invalid URL scheme '{parsed.scheme}'. Only http and https are allowed."
         )

backend/requirements/default.txt

@@ -698,7 +698,7 @@ py-key-value-aio==0.4.4
     # via fastmcp
 pyairtable==3.0.1
     # via onyx
-pyasn1==0.6.2
+pyasn1==0.6.3
     # via
     #   pyasn1-modules
     #   rsa
@@ -752,7 +752,7 @@ pypandoc-binary==1.16.2
     # via onyx
 pyparsing==3.2.5
     # via httplib2
-pypdf==6.8.0
+pypdf==6.9.1
     # via
     #   onyx
     #   unstructured-client

backend/requirements/dev.txt

@@ -263,7 +263,7 @@ oauthlib==3.2.2
     # via
     #   kubernetes
     #   requests-oauthlib
-onyx-devtools==0.7.0
+onyx-devtools==0.7.1
     # via onyx
 openai==2.14.0
     # via
@@ -326,7 +326,7 @@ pure-eval==0.2.3
     # via stack-data
 py==1.11.0
     # via retry
-pyasn1==0.6.2
+pyasn1==0.6.3
     # via
     #   pyasn1-modules
     #   rsa

backend/requirements/ee.txt

@@ -195,7 +195,7 @@ propcache==0.4.1
     #   yarl
 py==1.11.0
     # via retry
-pyasn1==0.6.2
+pyasn1==0.6.3
     # via
     #   pyasn1-modules
     #   rsa

backend/requirements/model_server.txt

@@ -285,7 +285,7 @@ psutil==7.1.3
     # via accelerate
 py==1.11.0
     # via retry
-pyasn1==0.6.2
+pyasn1==0.6.3
     # via
     #   pyasn1-modules
     #   rsa

backend/scripts/run_industryrag_bench_questions.py

@@ -0,0 +1,471 @@
+from __future__ import annotations
+
+import argparse
+import asyncio
+import json
+import logging
+import sys
+from dataclasses import asdict
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+from typing import TypedDict
+from typing import TypeGuard
+
+import aiohttp
+
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s %(levelname)s %(message)s",
+)
+logger = logging.getLogger(__name__)
+
+
+DEFAULT_API_BASE = "http://localhost:3000"
+INTERNAL_SEARCH_TOOL_NAME = "internal_search"
+INTERNAL_SEARCH_IN_CODE_TOOL_ID = "SearchTool"
+MAX_REQUEST_ATTEMPTS = 5
+RETRIABLE_STATUS_CODES = {429, 500, 502, 503, 504}
+
+
+@dataclass(frozen=True)
+class QuestionRecord:
+    question_id: str
+    question: str
+
+
+@dataclass(frozen=True)
+class AnswerRecord:
+    question_id: str
+    answer: str
+    document_ids: list[str]
+
+
+@dataclass(frozen=True)
+class FailedQuestionRecord:
+    question_id: str
+    error: str
+
+
+class Citation(TypedDict, total=False):
+    citation_number: int
+    document_id: str
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description=(
+            "Submit questions to Onyx chat with internal search forced and write "
+            "answers to a JSONL file."
+        )
+    )
+    parser.add_argument(
+        "--questions-file",
+        type=Path,
+        required=True,
+        help="Path to the input questions JSONL file.",
+    )
+    parser.add_argument(
+        "--output-file",
+        type=Path,
+        required=True,
+        help="Path to the output answers JSONL file.",
+    )
+    parser.add_argument(
+        "--api-key",
+        type=str,
+        required=True,
+        help="API key used to authenticate against Onyx.",
+    )
+    parser.add_argument(
+        "--api-base",
+        type=str,
+        default=DEFAULT_API_BASE,
+        help=(
+            "Frontend base URL for Onyx. If `/api` is omitted, it will be added "
+            f"automatically. Default: {DEFAULT_API_BASE}"
+        ),
+    )
+    parser.add_argument(
+        "--parallelism",
+        type=int,
+        default=1,
+        help="Number of questions to process in parallel. Default: 1.",
+    )
+    parser.add_argument(
+        "--max-questions",
+        type=int,
+        default=None,
+        help="Optional cap on how many questions to process. Defaults to all.",
+    )
+    return parser.parse_args()
+
+
+def normalize_api_base(api_base: str) -> str:
+    normalized = api_base.rstrip("/")
+    if normalized.endswith("/api"):
+        return normalized
+    return f"{normalized}/api"
+
+
+def load_questions(questions_file: Path) -> list[QuestionRecord]:
+    if not questions_file.exists():
+        raise FileNotFoundError(f"Questions file not found: {questions_file}")
+
+    questions: list[QuestionRecord] = []
+    with questions_file.open("r", encoding="utf-8") as file:
+        for line_number, line in enumerate(file, start=1):
+            stripped_line = line.strip()
+            if not stripped_line:
+                continue
+
+            try:
+                payload = json.loads(stripped_line)
+            except json.JSONDecodeError as exc:
+                raise ValueError(
+                    f"Invalid JSON on line {line_number} of {questions_file}"
+                ) from exc
+
+            question_id = payload.get("question_id")
+            question = payload.get("question")
+
+            if not isinstance(question_id, str) or not question_id:
+                raise ValueError(
+                    f"Line {line_number} is missing a non-empty `question_id`."
+                )
+            if not isinstance(question, str) or not question:
+                raise ValueError(
+                    f"Line {line_number} is missing a non-empty `question`."
+                )
+
+            questions.append(QuestionRecord(question_id=question_id, question=question))
+
+    return questions
+
+
+async def read_json_response(
+    response: aiohttp.ClientResponse,
+) -> dict[str, Any] | list[dict[str, Any]]:
+    response_text = await response.text()
+    if response.status >= 400:
+        raise RuntimeError(
+            f"Request to {response.url} failed with {response.status}: {response_text}"
+        )
+
+    try:
+        payload = json.loads(response_text)
+    except json.JSONDecodeError as exc:
+        raise RuntimeError(
+            f"Request to {response.url} returned non-JSON content: {response_text}"
+        ) from exc
+
+    if not isinstance(payload, (dict, list)):
+        raise RuntimeError(
+            f"Unexpected response payload type from {response.url}: {type(payload)}"
+        )
+
+    return payload
+
+
+async def request_json_with_retries(
+    session: aiohttp.ClientSession,
+    method: str,
+    url: str,
+    headers: dict[str, str],
+    json_payload: dict[str, Any] | None = None,
+) -> dict[str, Any] | list[dict[str, Any]]:
+    backoff_seconds = 1.0
+
+    for attempt in range(1, MAX_REQUEST_ATTEMPTS + 1):
+        try:
+            async with session.request(
+                method=method,
+                url=url,
+                headers=headers,
+                json=json_payload,
+            ) as response:
+                if (
+                    response.status in RETRIABLE_STATUS_CODES
+                    and attempt < MAX_REQUEST_ATTEMPTS
+                ):
+                    response_text = await response.text()
+                    logger.warning(
+                        "Retryable response from %s on attempt %s/%s: %s %s",
+                        url,
+                        attempt,
+                        MAX_REQUEST_ATTEMPTS,
+                        response.status,
+                        response_text,
+                    )
+                    await asyncio.sleep(backoff_seconds)
+                    backoff_seconds *= 2
+                    continue
+
+                return await read_json_response(response)
+        except (aiohttp.ClientError, asyncio.TimeoutError) as exc:
+            if attempt == MAX_REQUEST_ATTEMPTS:
+                raise RuntimeError(
+                    f"Request to {url} failed after {MAX_REQUEST_ATTEMPTS} attempts."
+                ) from exc
+
+            logger.warning(
+                "Request to %s failed on attempt %s/%s: %s",
+                url,
+                attempt,
+                MAX_REQUEST_ATTEMPTS,
+                exc,
+            )
+            await asyncio.sleep(backoff_seconds)
+            backoff_seconds *= 2
+
+    raise RuntimeError(f"Request to {url} failed unexpectedly.")
+
+
+def extract_document_ids(citation_info: object) -> list[str]:
+    if not isinstance(citation_info, list):
+        return []
+
+    sorted_citations = sorted(
+        (citation for citation in citation_info if _is_valid_citation(citation)),
+        key=_citation_sort_key,
+    )
+
+    document_ids: list[str] = []
+    seen_document_ids: set[str] = set()
+    for citation in sorted_citations:
+        document_id = citation["document_id"]
+        if document_id not in seen_document_ids:
+            seen_document_ids.add(document_id)
+            document_ids.append(document_id)
+
+    return document_ids
+
+
+def _is_valid_citation(citation: object) -> TypeGuard[Citation]:
+    return (
+        isinstance(citation, dict)
+        and isinstance(citation.get("document_id"), str)
+        and bool(citation["document_id"])
+    )
+
+
+def _citation_sort_key(citation: Citation) -> int:
+    citation_number = citation.get("citation_number")
+    if isinstance(citation_number, int):
+        return citation_number
+    return sys.maxsize
+
+
+async def fetch_internal_search_tool_id(
+    session: aiohttp.ClientSession,
+    api_base: str,
+    headers: dict[str, str],
+) -> int:
+    payload = await request_json_with_retries(
+        session=session,
+        method="GET",
+        url=f"{api_base}/tool",
+        headers=headers,
+    )
+
+    if not isinstance(payload, list):
+        raise RuntimeError("Expected `/tool` to return a list.")
+
+    for tool in payload:
+        if not isinstance(tool, dict):
+            continue
+
+        if tool.get("in_code_tool_id") == INTERNAL_SEARCH_IN_CODE_TOOL_ID:
+            tool_id = tool.get("id")
+            if isinstance(tool_id, int):
+                return tool_id
+
+    for tool in payload:
+        if not isinstance(tool, dict):
+            continue
+
+        if tool.get("name") == INTERNAL_SEARCH_TOOL_NAME:
+            tool_id = tool.get("id")
+            if isinstance(tool_id, int):
+                return tool_id
+
+    raise RuntimeError(
+        "Could not find the internal search tool in `/tool`. "
+        "Make sure SearchTool is available for this environment."
+    )
+
+
+async def submit_question(
+    session: aiohttp.ClientSession,
+    api_base: str,
+    headers: dict[str, str],
+    internal_search_tool_id: int,
+    question_record: QuestionRecord,
+) -> AnswerRecord:
+    payload = {
+        "message": question_record.question,
+        "chat_session_info": {"persona_id": 0},
+        "parent_message_id": None,
+        "file_descriptors": [],
+        "allowed_tool_ids": [internal_search_tool_id],
+        "forced_tool_id": internal_search_tool_id,
+        "stream": False,
+    }
+
+    response_payload = await request_json_with_retries(
+        session=session,
+        method="POST",
+        url=f"{api_base}/chat/send-chat-message",
+        headers=headers,
+        json_payload=payload,
+    )
+
+    if not isinstance(response_payload, dict):
+        raise RuntimeError(
+            "Expected `/chat/send-chat-message` to return an object when `stream=false`."
+        )
+
+    answer = response_payload.get("answer_citationless")
+    if not isinstance(answer, str):
+        answer = response_payload.get("answer")
+
+    if not isinstance(answer, str):
+        raise RuntimeError(
+            f"Response for question {question_record.question_id} is missing `answer`."
+        )
+
+    return AnswerRecord(
+        question_id=question_record.question_id,
+        answer=answer,
+        document_ids=extract_document_ids(response_payload.get("citation_info")),
+    )
+
+
+async def generate_answers(
+    questions: list[QuestionRecord],
+    output_file: Path,
+    api_base: str,
+    api_key: str,
+    parallelism: int,
+) -> None:
+    if parallelism < 1:
+        raise ValueError("`--parallelism` must be at least 1.")
+
+    headers = {
+        "Authorization": f"Bearer {api_key}",
+        "Content-Type": "application/json",
+    }
+
+    timeout = aiohttp.ClientTimeout(
+        total=None,
+        connect=30,
+        sock_connect=30,
+        sock_read=600,
+    )
+    connector = aiohttp.TCPConnector(limit=parallelism)
+
+    output_file.parent.mkdir(parents=True, exist_ok=True)
+    with output_file.open("a", encoding="utf-8") as file:
+        async with aiohttp.ClientSession(
+            timeout=timeout, connector=connector
+        ) as session:
+            internal_search_tool_id = await fetch_internal_search_tool_id(
+                session=session,
+                api_base=api_base,
+                headers=headers,
+            )
+            logger.info("Using internal search tool id %s", internal_search_tool_id)
+
+            semaphore = asyncio.Semaphore(parallelism)
+            progress_lock = asyncio.Lock()
+            write_lock = asyncio.Lock()
+            completed = 0
+            successful = 0
+            failed_questions: list[FailedQuestionRecord] = []
+            total = len(questions)
+
+            async def process_question(question_record: QuestionRecord) -> None:
+                nonlocal completed
+                nonlocal successful
+
+                try:
+                    async with semaphore:
+                        result = await submit_question(
+                            session=session,
+                            api_base=api_base,
+                            headers=headers,
+                            internal_search_tool_id=internal_search_tool_id,
+                            question_record=question_record,
+                        )
+                except Exception as exc:
+                    async with progress_lock:
+                        completed += 1
+                        failed_questions.append(
+                            FailedQuestionRecord(
+                                question_id=question_record.question_id,
+                                error=str(exc),
+                            )
+                        )
+                        logger.exception(
+                            "Failed question %s (%s/%s)",
+                            question_record.question_id,
+                            completed,
+                            total,
+                        )
+                    return
+
+                async with write_lock:
+                    file.write(json.dumps(asdict(result), ensure_ascii=False))
+                    file.write("\n")
+                    file.flush()
+
+                async with progress_lock:
+                    completed += 1
+                    successful += 1
+                    logger.info("Processed %s/%s questions", completed, total)
+
+            await asyncio.gather(
+                *(process_question(question_record) for question_record in questions)
+            )
+
+            if failed_questions:
+                logger.warning(
+                    "Completed with %s failed questions and %s successful questions.",
+                    len(failed_questions),
+                    successful,
+                )
+                for failed_question in failed_questions:
+                    logger.warning(
+                        "Failed question %s: %s",
+                        failed_question.question_id,
+                        failed_question.error,
+                    )
+
+
+def main() -> None:
+    args = parse_args()
+    questions = load_questions(args.questions_file)
+    api_base = normalize_api_base(args.api_base)
+
+    if args.max_questions is not None:
+        if args.max_questions < 1:
+            raise ValueError("`--max-questions` must be at least 1 when provided.")
+        questions = questions[: args.max_questions]
+
+    logger.info("Loaded %s questions from %s", len(questions), args.questions_file)
+    logger.info("Writing answers to %s", args.output_file)
+
+    asyncio.run(
+        generate_answers(
+            questions=questions,
+            output_file=args.output_file,
+            api_base=api_base,
+            api_key=args.api_key,
+            parallelism=args.parallelism,
+        )
+    )
+
+
+if __name__ == "__main__":
+    main()

backend/scripts/upload_files_as_connectors.py

@@ -0,0 +1,291 @@
+"""
+Script to upload files from a directory as individual file connectors in Onyx.
+Each file gets its own connector named after the file.
+
+Usage:
+    python upload_files_as_connectors.py --data-dir /path/to/files --api-key YOUR_KEY
+    python upload_files_as_connectors.py --data-dir /path/to/files --api-key YOUR_KEY --api-base http://onyxserver:3000
+    python upload_files_as_connectors.py --data-dir /path/to/files --api-key YOUR_KEY --file-glob '*.zip'
+
+Requires:
+    pip install requests
+"""
+
+import argparse
+import fnmatch
+import os
+import sys
+import threading
+import time
+
+import requests
+
+REQUEST_TIMEOUT = 900  # 15 minutes
+
+
+def _elapsed_printer(label: str, stop_event: threading.Event) -> None:
+    """Print a live elapsed-time counter until stop_event is set."""
+    start = time.monotonic()
+    while not stop_event.wait(timeout=1):
+        elapsed = int(time.monotonic() - start)
+        m, s = divmod(elapsed, 60)
+        print(f"\r  {label} ... {m:02d}:{s:02d}", end="", flush=True)
+    elapsed = int(time.monotonic() - start)
+    m, s = divmod(elapsed, 60)
+    print(f"\r  {label} ... {m:02d}:{s:02d} done")
+
+
+def _timed_request(label: str, fn: object) -> requests.Response:
+    """Run a request function while displaying a live elapsed timer."""
+    stop = threading.Event()
+    t = threading.Thread(target=_elapsed_printer, args=(label, stop), daemon=True)
+    t.start()
+    try:
+        resp = fn()  # type: ignore[operator]
+    finally:
+        stop.set()
+        t.join()
+    return resp
+
+
+def upload_file(
+    session: requests.Session, base_url: str, file_path: str
+) -> dict | None:
+    """Upload a single file and return the response with file_paths and file_names."""
+    with open(file_path, "rb") as f:
+        resp = _timed_request(
+            "Uploading",
+            lambda: session.post(
+                f"{base_url}/api/manage/admin/connector/file/upload",
+                files={"files": (os.path.basename(file_path), f)},
+                timeout=REQUEST_TIMEOUT,
+            ),
+        )
+    if not resp.ok:
+        print(f"  ERROR uploading: {resp.text}")
+        return None
+    return resp.json()
+
+
+def create_connector(
+    session: requests.Session,
+    base_url: str,
+    name: str,
+    file_paths: list[str],
+    file_names: list[str],
+    zip_metadata_file_id: str | None,
+) -> int | None:
+    """Create a file connector and return its ID."""
+    resp = _timed_request(
+        "Creating connector",
+        lambda: session.post(
+            f"{base_url}/api/manage/admin/connector",
+            json={
+                "name": name,
+                "source": "file",
+                "input_type": "load_state",
+                "connector_specific_config": {
+                    "file_locations": file_paths,
+                    "file_names": file_names,
+                    "zip_metadata_file_id": zip_metadata_file_id,
+                },
+                "refresh_freq": None,
+                "prune_freq": None,
+                "indexing_start": None,
+                "access_type": "public",
+                "groups": [],
+            },
+            timeout=REQUEST_TIMEOUT,
+        ),
+    )
+    if not resp.ok:
+        print(f"  ERROR creating connector: {resp.text}")
+        return None
+    return resp.json()["id"]
+
+
+def create_credential(
+    session: requests.Session, base_url: str, name: str
+) -> int | None:
+    """Create a dummy credential for the file connector."""
+    resp = session.post(
+        f"{base_url}/api/manage/credential",
+        json={
+            "credential_json": {},
+            "admin_public": True,
+            "source": "file",
+            "curator_public": True,
+            "groups": [],
+            "name": name,
+        },
+        timeout=REQUEST_TIMEOUT,
+    )
+    if not resp.ok:
+        print(f"  ERROR creating credential: {resp.text}")
+        return None
+    return resp.json()["id"]
+
+
+def link_credential(
+    session: requests.Session,
+    base_url: str,
+    connector_id: int,
+    credential_id: int,
+    name: str,
+) -> bool:
+    """Link the connector to the credential (create CC pair)."""
+    resp = session.put(
+        f"{base_url}/api/manage/connector/{connector_id}/credential/{credential_id}",
+        json={
+            "name": name,
+            "access_type": "public",
+            "groups": [],
+            "auto_sync_options": None,
+            "processing_mode": "REGULAR",
+        },
+        timeout=REQUEST_TIMEOUT,
+    )
+    if not resp.ok:
+        print(f"  ERROR linking credential: {resp.text}")
+        return False
+    return True
+
+
+def run_connector(
+    session: requests.Session,
+    base_url: str,
+    connector_id: int,
+    credential_id: int,
+) -> bool:
+    """Trigger the connector to start indexing."""
+    resp = session.post(
+        f"{base_url}/api/manage/admin/connector/run-once",
+        json={
+            "connector_id": connector_id,
+            "credentialIds": [credential_id],
+            "from_beginning": False,
+        },
+        timeout=REQUEST_TIMEOUT,
+    )
+    if not resp.ok:
+        print(f"  ERROR running connector: {resp.text}")
+        return False
+    return True
+
+
+def process_file(session: requests.Session, base_url: str, file_path: str) -> bool:
+    """Process a single file through the full connector creation flow."""
+    file_name = os.path.basename(file_path)
+    connector_name = file_name
+    print(f"Processing: {file_name}")
+
+    # Step 1: Upload
+    upload_resp = upload_file(session, base_url, file_path)
+    if not upload_resp:
+        return False
+
+    # Step 2: Create connector
+    connector_id = create_connector(
+        session,
+        base_url,
+        name=f"FileConnector-{connector_name}",
+        file_paths=upload_resp["file_paths"],
+        file_names=upload_resp["file_names"],
+        zip_metadata_file_id=upload_resp.get("zip_metadata_file_id"),
+    )
+    if connector_id is None:
+        return False
+
+    # Step 3: Create credential
+    credential_id = create_credential(session, base_url, name=connector_name)
+    if credential_id is None:
+        return False
+
+    # Step 4: Link connector to credential
+    if not link_credential(
+        session, base_url, connector_id, credential_id, connector_name
+    ):
+        return False
+
+    # Step 5: Trigger indexing
+    if not run_connector(session, base_url, connector_id, credential_id):
+        return False
+
+    print(f"  OK (connector_id={connector_id})")
+    return True
+
+
+def get_authenticated_session(api_key: str) -> requests.Session:
+    """Create a session authenticated with an API key."""
+    session = requests.Session()
+    session.headers.update({"Authorization": f"Bearer {api_key}"})
+    return session
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(
+        description="Upload files as individual Onyx file connectors."
+    )
+    parser.add_argument(
+        "--data-dir",
+        required=True,
+        help="Directory containing files to upload.",
+    )
+    parser.add_argument(
+        "--api-base",
+        default="http://localhost:3000",
+        help="Base URL for the Onyx API (default: http://localhost:3000).",
+    )
+    parser.add_argument(
+        "--api-key",
+        required=True,
+        help="API key for authentication.",
+    )
+    parser.add_argument(
+        "--file-glob",
+        default=None,
+        help="Glob pattern to filter files (e.g. '*.json', '*.zip').",
+    )
+    args = parser.parse_args()
+
+    data_dir = args.data_dir
+    base_url = args.api_base.rstrip("/")
+    api_key = args.api_key
+    file_glob = args.file_glob
+
+    if not os.path.isdir(data_dir):
+        print(f"Error: {data_dir} is not a directory")
+        sys.exit(1)
+
+    script_path = os.path.realpath(__file__)
+    files = sorted(
+        os.path.join(data_dir, f)
+        for f in os.listdir(data_dir)
+        if os.path.isfile(os.path.join(data_dir, f))
+        and os.path.realpath(os.path.join(data_dir, f)) != script_path
+        and (file_glob is None or fnmatch.fnmatch(f, file_glob))
+    )
+
+    if not files:
+        print(f"No files found in {data_dir}")
+        sys.exit(1)
+
+    print(f"Found {len(files)} file(s) in {data_dir}\n")
+
+    session = get_authenticated_session(api_key)
+
+    success = 0
+    failed = 0
+    for file_path in files:
+        if process_file(session, base_url, file_path):
+            success += 1
+        else:
+            failed += 1
+        # Small delay to avoid overwhelming the server
+        time.sleep(0.5)
+
+    print(f"\nDone: {success} succeeded, {failed} failed out of {len(files)} files.")
+
+
+if __name__ == "__main__":
+    main()

backend/tests/external_dependency_unit/celery/test_user_file_delete_queue.py

@@ -0,0 +1,274 @@
+"""
+External dependency unit tests for user file delete queue protections.
+
+Verifies that the three mechanisms added to check_for_user_file_delete work
+correctly:
+
+1. Queue depth backpressure – when the broker queue exceeds
+   USER_FILE_DELETE_MAX_QUEUE_DEPTH, no new tasks are enqueued.
+
+2. Per-file Redis guard key – if the guard key for a file already exists in
+   Redis, that file is skipped even though it is still in DELETING status.
+
+3. Task expiry – every send_task call carries expires=
+   CELERY_USER_FILE_DELETE_TASK_EXPIRES so that stale queued tasks are
+   discarded by workers automatically.
+
+Also verifies that delete_user_file_impl clears the guard key the moment
+it is picked up by a worker.
+
+Uses real Redis (DB 0 via get_redis_client) and real PostgreSQL for UserFile
+rows.  The Celery app is provided as a MagicMock injected via a PropertyMock
+on the task class so no real broker is needed.
+"""
+
+from collections.abc import Generator
+from contextlib import contextmanager
+from typing import Any
+from unittest.mock import MagicMock
+from unittest.mock import patch
+from unittest.mock import PropertyMock
+from uuid import uuid4
+
+from sqlalchemy.orm import Session
+
+from onyx.background.celery.tasks.user_file_processing.tasks import (
+    _user_file_delete_lock_key,
+)
+from onyx.background.celery.tasks.user_file_processing.tasks import (
+    _user_file_delete_queued_key,
+)
+from onyx.background.celery.tasks.user_file_processing.tasks import (
+    check_for_user_file_delete,
+)
+from onyx.background.celery.tasks.user_file_processing.tasks import (
+    process_single_user_file_delete,
+)
+from onyx.configs.constants import CELERY_USER_FILE_DELETE_TASK_EXPIRES
+from onyx.configs.constants import OnyxCeleryQueues
+from onyx.configs.constants import OnyxCeleryTask
+from onyx.configs.constants import USER_FILE_DELETE_MAX_QUEUE_DEPTH
+from onyx.db.enums import UserFileStatus
+from onyx.db.models import UserFile
+from onyx.redis.redis_pool import get_redis_client
+from tests.external_dependency_unit.conftest import create_test_user
+from tests.external_dependency_unit.constants import TEST_TENANT_ID
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+_PATCH_QUEUE_LEN = (
+    "onyx.background.celery.tasks.user_file_processing.tasks.celery_get_queue_length"
+)
+
+
+def _create_deleting_user_file(db_session: Session, user_id: object) -> UserFile:
+    """Insert a UserFile in DELETING status and return it."""
+    uf = UserFile(
+        id=uuid4(),
+        user_id=user_id,
+        file_id=f"test_file_{uuid4().hex[:8]}",
+        name=f"test_{uuid4().hex[:8]}.txt",
+        file_type="text/plain",
+        status=UserFileStatus.DELETING,
+    )
+    db_session.add(uf)
+    db_session.commit()
+    db_session.refresh(uf)
+    return uf
+
+
+@contextmanager
+def _patch_task_app(task: Any, mock_app: MagicMock) -> Generator[None, None, None]:
+    """Patch the ``app`` property on *task*'s class so that ``self.app``
+    inside the task function returns *mock_app*.
+
+    With ``bind=True``, ``task.run`` is a bound method whose ``__self__`` is
+    the actual task instance.  We patch ``app`` on that instance's class
+    (a unique Celery-generated Task subclass) so the mock is scoped to this
+    task only.
+    """
+    task_instance = task.run.__self__
+    with patch.object(
+        type(task_instance), "app", new_callable=PropertyMock, return_value=mock_app
+    ):
+        yield
+
+
+# ---------------------------------------------------------------------------
+# Test classes
+# ---------------------------------------------------------------------------
+
+
+class TestDeleteQueueDepthBackpressure:
+    """Protection 1: skip all enqueuing when the broker queue is too deep."""
+
+    def test_no_tasks_enqueued_when_queue_over_limit(
+        self,
+        db_session: Session,
+        tenant_context: None,  # noqa: ARG002
+    ) -> None:
+        """When the queue depth exceeds the limit the beat cycle is skipped."""
+        user = create_test_user(db_session, "del_bp_user")
+        _create_deleting_user_file(db_session, user.id)
+
+        mock_app = MagicMock()
+
+        with (
+            _patch_task_app(check_for_user_file_delete, mock_app),
+            patch(_PATCH_QUEUE_LEN, return_value=USER_FILE_DELETE_MAX_QUEUE_DEPTH + 1),
+        ):
+            check_for_user_file_delete.run(tenant_id=TEST_TENANT_ID)
+
+        mock_app.send_task.assert_not_called()
+
+
+class TestDeletePerFileGuardKey:
+    """Protection 2: per-file Redis guard key prevents duplicate enqueue."""
+
+    def test_guarded_file_not_re_enqueued(
+        self,
+        db_session: Session,
+        tenant_context: None,  # noqa: ARG002
+    ) -> None:
+        """A file whose guard key is already set in Redis is skipped."""
+        user = create_test_user(db_session, "del_guard_user")
+        uf = _create_deleting_user_file(db_session, user.id)
+
+        redis_client = get_redis_client(tenant_id=TEST_TENANT_ID)
+        guard_key = _user_file_delete_queued_key(uf.id)
+        redis_client.setex(guard_key, CELERY_USER_FILE_DELETE_TASK_EXPIRES, 1)
+
+        mock_app = MagicMock()
+
+        try:
+            with (
+                _patch_task_app(check_for_user_file_delete, mock_app),
+                patch(_PATCH_QUEUE_LEN, return_value=0),
+            ):
+                check_for_user_file_delete.run(tenant_id=TEST_TENANT_ID)
+
+            # send_task must not have been called with this specific file's ID
+            for call in mock_app.send_task.call_args_list:
+                kwargs = call.kwargs.get("kwargs", {})
+                assert kwargs.get("user_file_id") != str(
+                    uf.id
+                ), f"File {uf.id} should have been skipped because its guard key exists"
+        finally:
+            redis_client.delete(guard_key)
+
+    def test_guard_key_exists_in_redis_after_enqueue(
+        self,
+        db_session: Session,
+        tenant_context: None,  # noqa: ARG002
+    ) -> None:
+        """After a file is enqueued its guard key is present in Redis with a TTL."""
+        user = create_test_user(db_session, "del_guard_set_user")
+        uf = _create_deleting_user_file(db_session, user.id)
+
+        redis_client = get_redis_client(tenant_id=TEST_TENANT_ID)
+        guard_key = _user_file_delete_queued_key(uf.id)
+        redis_client.delete(guard_key)  # clean slate
+
+        mock_app = MagicMock()
+
+        try:
+            with (
+                _patch_task_app(check_for_user_file_delete, mock_app),
+                patch(_PATCH_QUEUE_LEN, return_value=0),
+            ):
+                check_for_user_file_delete.run(tenant_id=TEST_TENANT_ID)
+
+            assert redis_client.exists(
+                guard_key
+            ), "Guard key should be set in Redis after enqueue"
+            ttl = int(redis_client.ttl(guard_key))  # type: ignore[arg-type]
+            assert (
+                0 < ttl <= CELERY_USER_FILE_DELETE_TASK_EXPIRES
+            ), f"Guard key TTL {ttl}s is outside the expected range (0, {CELERY_USER_FILE_DELETE_TASK_EXPIRES}]"
+        finally:
+            redis_client.delete(guard_key)
+
+
+class TestDeleteTaskExpiry:
+    """Protection 3: every send_task call includes an expires value."""
+
+    def test_send_task_called_with_expires(
+        self,
+        db_session: Session,
+        tenant_context: None,  # noqa: ARG002
+    ) -> None:
+        """send_task is called with the correct queue, task name, and expires."""
+        user = create_test_user(db_session, "del_expires_user")
+        uf = _create_deleting_user_file(db_session, user.id)
+
+        redis_client = get_redis_client(tenant_id=TEST_TENANT_ID)
+        guard_key = _user_file_delete_queued_key(uf.id)
+        redis_client.delete(guard_key)
+
+        mock_app = MagicMock()
+
+        try:
+            with (
+                _patch_task_app(check_for_user_file_delete, mock_app),
+                patch(_PATCH_QUEUE_LEN, return_value=0),
+            ):
+                check_for_user_file_delete.run(tenant_id=TEST_TENANT_ID)
+
+            # At least one task should have been submitted (for our file)
+            assert (
+                mock_app.send_task.call_count >= 1
+            ), "Expected at least one task to be submitted"
+
+            # Every submitted task must carry expires
+            for call in mock_app.send_task.call_args_list:
+                assert call.args[0] == OnyxCeleryTask.DELETE_SINGLE_USER_FILE
+                assert call.kwargs.get("queue") == OnyxCeleryQueues.USER_FILE_DELETE
+                assert (
+                    call.kwargs.get("expires") == CELERY_USER_FILE_DELETE_TASK_EXPIRES
+                ), "Task must be submitted with the correct expires value to prevent stale task accumulation"
+        finally:
+            redis_client.delete(guard_key)
+
+
+class TestDeleteWorkerClearsGuardKey:
+    """process_single_user_file_delete removes the guard key when it picks up a task."""
+
+    def test_guard_key_deleted_on_pickup(
+        self,
+        tenant_context: None,  # noqa: ARG002
+    ) -> None:
+        """The guard key is deleted before the worker does any real work.
+
+        We simulate an already-locked file so delete_user_file_impl returns
+        early – but crucially, after the guard key deletion.
+        """
+        user_file_id = str(uuid4())
+
+        redis_client = get_redis_client(tenant_id=TEST_TENANT_ID)
+        guard_key = _user_file_delete_queued_key(user_file_id)
+
+        # Simulate the guard key set when the beat enqueued the task
+        redis_client.setex(guard_key, CELERY_USER_FILE_DELETE_TASK_EXPIRES, 1)
+        assert redis_client.exists(guard_key), "Guard key must exist before pickup"
+
+        # Hold the per-file delete lock so the worker exits early without
+        # touching the database or file store.
+        lock_key = _user_file_delete_lock_key(user_file_id)
+        delete_lock = redis_client.lock(lock_key, timeout=10)
+        acquired = delete_lock.acquire(blocking=False)
+        assert acquired, "Should be able to acquire the delete lock for this test"
+
+        try:
+            process_single_user_file_delete.run(
+                user_file_id=user_file_id,
+                tenant_id=TEST_TENANT_ID,
+            )
+        finally:
+            if delete_lock.owned():
+                delete_lock.release()
+
+        assert not redis_client.exists(
+            guard_key
+        ), "Guard key should be deleted when the worker picks up the task"

backend/tests/external_dependency_unit/document_index/test_document_index_old.py

@@ -297,6 +297,10 @@ def index_batch_params(
 class TestDocumentIndexOld:
     """Tests the old DocumentIndex interface."""
 
+    # TODO(ENG-3864)(andrei): Re-enable this test.
+    @pytest.mark.xfail(
+        reason="Flaky test: Retrieved chunks vary non-deterministically before and after changing user projects and personas. Likely a timing issue with the index being updated."
+    )
     def test_update_single_can_clear_user_projects_and_personas(
         self,
         document_indices: list[DocumentIndex],

backend/tests/external_dependency_unit/opensearch/test_opensearch_client.py

@@ -22,18 +22,26 @@ from onyx.document_index.interfaces_new import TenantState
 from onyx.document_index.opensearch.client import OpenSearchIndexClient
 from onyx.document_index.opensearch.client import wait_for_opensearch_with_timeout
 from onyx.document_index.opensearch.constants import DEFAULT_MAX_CHUNK_SIZE
+from onyx.document_index.opensearch.constants import HybridSearchNormalizationPipeline
+from onyx.document_index.opensearch.constants import HybridSearchSubqueryConfiguration
 from onyx.document_index.opensearch.opensearch_document_index import (
     generate_opensearch_filtered_access_control_list,
 )
 from onyx.document_index.opensearch.schema import CONTENT_FIELD_NAME
 from onyx.document_index.opensearch.schema import DocumentChunk
+from onyx.document_index.opensearch.schema import DocumentChunkWithoutVectors
 from onyx.document_index.opensearch.schema import DocumentSchema
 from onyx.document_index.opensearch.schema import get_opensearch_doc_chunk_id
 from onyx.document_index.opensearch.search import DocumentQuery
 from onyx.document_index.opensearch.search import (
-    MIN_MAX_NORMALIZATION_PIPELINE_CONFIG,
+    get_min_max_normalization_pipeline_name_and_config,
+)
+from onyx.document_index.opensearch.search import (
+    get_normalization_pipeline_name_and_config,
+)
+from onyx.document_index.opensearch.search import (
+    get_zscore_normalization_pipeline_name_and_config,
 )
-from onyx.document_index.opensearch.search import MIN_MAX_NORMALIZATION_PIPELINE_NAME
 from shared_configs.configs import POSTGRES_DEFAULT_SCHEMA
 
 
@@ -49,6 +57,63 @@ def _patch_global_tenant_state(monkeypatch: pytest.MonkeyPatch, state: bool) ->
     monkeypatch.setattr("onyx.document_index.opensearch.schema.MULTI_TENANT", state)
 
 
+def _patch_hybrid_search_subquery_configuration(
+    monkeypatch: pytest.MonkeyPatch, configuration: HybridSearchSubqueryConfiguration
+) -> None:
+    """
+    Patches HYBRID_SEARCH_SUBQUERY_CONFIGURATION wherever necessary for this
+    test file.
+
+    Args:
+        monkeypatch: The test instance's monkeypatch instance, used for
+            patching.
+        configuration: The intended state of
+            HYBRID_SEARCH_SUBQUERY_CONFIGURATION.
+    """
+    monkeypatch.setattr(
+        "onyx.document_index.opensearch.constants.HYBRID_SEARCH_SUBQUERY_CONFIGURATION",
+        configuration,
+    )
+    monkeypatch.setattr(
+        "onyx.document_index.opensearch.search.HYBRID_SEARCH_SUBQUERY_CONFIGURATION",
+        configuration,
+    )
+
+
+def _patch_hybrid_search_normalization_pipeline(
+    monkeypatch: pytest.MonkeyPatch, pipeline: HybridSearchNormalizationPipeline
+) -> None:
+    """
+    Patches HYBRID_SEARCH_NORMALIZATION_PIPELINE wherever necessary for this
+    test file.
+    """
+    monkeypatch.setattr(
+        "onyx.document_index.opensearch.constants.HYBRID_SEARCH_NORMALIZATION_PIPELINE",
+        pipeline,
+    )
+    monkeypatch.setattr(
+        "onyx.document_index.opensearch.search.HYBRID_SEARCH_NORMALIZATION_PIPELINE",
+        pipeline,
+    )
+
+
+def _patch_opensearch_match_highlights_disabled(
+    monkeypatch: pytest.MonkeyPatch, disabled: bool
+) -> None:
+    """
+    Patches OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED wherever necessary for this
+    test file.
+    """
+    monkeypatch.setattr(
+        "onyx.configs.app_configs.OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED",
+        disabled,
+    )
+    monkeypatch.setattr(
+        "onyx.document_index.opensearch.search.OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED",
+        disabled,
+    )
+
+
 def _create_test_document_chunk(
     document_id: str,
     content: str,
@@ -144,14 +209,27 @@ def test_client(
 @pytest.fixture(scope="function")
 def search_pipeline(test_client: OpenSearchIndexClient) -> Generator[None, None, None]:
     """Creates a search pipeline for testing with automatic cleanup."""
+    min_max_normalization_pipeline_name, min_max_normalization_pipeline_config = (
+        get_min_max_normalization_pipeline_name_and_config()
+    )
+    zscore_normalization_pipeline_name, zscore_normalization_pipeline_config = (
+        get_zscore_normalization_pipeline_name_and_config()
+    )
     test_client.create_search_pipeline(
-        pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME,
-        pipeline_body=MIN_MAX_NORMALIZATION_PIPELINE_CONFIG,
+        pipeline_id=min_max_normalization_pipeline_name,
+        pipeline_body=min_max_normalization_pipeline_config,
+    )
+    test_client.create_search_pipeline(
+        pipeline_id=zscore_normalization_pipeline_name,
+        pipeline_body=zscore_normalization_pipeline_config,
     )
     yield  # Test runs here.
     try:
         test_client.delete_search_pipeline(
-            pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME,
+            pipeline_id=min_max_normalization_pipeline_name,
+        )
+        test_client.delete_search_pipeline(
+            pipeline_id=zscore_normalization_pipeline_name,
         )
     except Exception:
         pass
@@ -166,7 +244,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=True
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
 
         # Under test.
         # Should not raise.
@@ -182,7 +260,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=True
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Under test.
@@ -211,7 +289,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=True
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
 
         test_client.create_index(mappings=mappings, settings=settings)
 
@@ -225,7 +303,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=True
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
 
         # Under test and postcondition.
         # Should return False before creation.
@@ -245,7 +323,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=True
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Under test.
@@ -280,7 +358,7 @@ class TestOpenSearchClient:
                 },
             },
         }
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=initial_mappings, settings=settings)
 
         # Under test.
@@ -323,7 +401,7 @@ class TestOpenSearchClient:
                 "test_field": {"type": "keyword"},
             },
         }
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=initial_mappings, settings=settings)
 
         # Under test and postcondition.
@@ -358,7 +436,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=True
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         # Create once - should succeed.
         test_client.create_index(mappings=mappings, settings=settings)
 
@@ -377,18 +455,19 @@ class TestOpenSearchClient:
         self, test_client: OpenSearchIndexClient
     ) -> None:
         """Tests creating and deleting a search pipeline."""
+        # Precondition.
+        pipeline_name, pipeline_config = get_normalization_pipeline_name_and_config()
+
         # Under test and postcondition.
         # Should not raise.
         test_client.create_search_pipeline(
-            pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME,
-            pipeline_body=MIN_MAX_NORMALIZATION_PIPELINE_CONFIG,
+            pipeline_id=pipeline_name,
+            pipeline_body=pipeline_config,
         )
 
         # Under test and postcondition.
         # Should not raise.
-        test_client.delete_search_pipeline(
-            pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME
-        )
+        test_client.delete_search_pipeline(pipeline_id=pipeline_name)
 
     def test_index_document(
         self, test_client: OpenSearchIndexClient, monkeypatch: pytest.MonkeyPatch
@@ -400,7 +479,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         doc = _create_test_document_chunk(
@@ -428,7 +507,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         docs = [
@@ -459,7 +538,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         doc = _create_test_document_chunk(
@@ -487,7 +566,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         original_doc = _create_test_document_chunk(
@@ -522,7 +601,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=False
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Under test and postcondition.
@@ -541,7 +620,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         doc = _create_test_document_chunk(
@@ -577,7 +656,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Under test.
@@ -598,7 +677,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Index multiple documents.
@@ -674,7 +753,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Create a document to update.
@@ -723,7 +802,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Under test and postcondition.
@@ -734,22 +813,22 @@ class TestOpenSearchClient:
                 properties_to_update={"hidden": True},
             )
 
-    def test_hybrid_search_with_pipeline(
+    def test_hybrid_search_configurations_and_pipelines(
         self,
         test_client: OpenSearchIndexClient,
         search_pipeline: None,  # noqa: ARG002
         monkeypatch: pytest.MonkeyPatch,
     ) -> None:
-        """Tests hybrid search with a normalization pipeline."""
+        """Tests all hybrid search configurations and pipelines."""
         # Precondition.
         _patch_global_tenant_state(monkeypatch, False)
+        _patch_opensearch_match_highlights_disabled(monkeypatch, False)
         tenant_state = TenantState(tenant_id=POSTGRES_DEFAULT_SCHEMA, multitenant=False)
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
-
         # Index documents.
         docs = {
             "doc-1": _create_test_document_chunk(
@@ -780,40 +859,62 @@ class TestOpenSearchClient:
         # Refresh index to make documents searchable.
         test_client.refresh_index()
 
-        # Search query.
-        query_text = "Python programming"
-        query_vector = _generate_test_vector(0.12)
-        search_body = DocumentQuery.get_hybrid_search_query(
-            query_text=query_text,
-            query_vector=query_vector,
-            num_hits=5,
-            tenant_state=tenant_state,
-            # We're not worried about filtering here. tenant_id in this object
-            # is not relevant.
-            index_filters=IndexFilters(access_control_list=None, tenant_id=None),
-            include_hidden=False,
-        )
+        for configuration in HybridSearchSubqueryConfiguration:
+            _patch_hybrid_search_subquery_configuration(monkeypatch, configuration)
+            for pipeline in HybridSearchNormalizationPipeline:
+                _patch_hybrid_search_normalization_pipeline(monkeypatch, pipeline)
+                pipeline_name, pipeline_config = (
+                    get_normalization_pipeline_name_and_config()
+                )
+                test_client.create_search_pipeline(
+                    pipeline_id=pipeline_name,
+                    pipeline_body=pipeline_config,
+                )
 
-        # Under test.
-        results = test_client.search(
-            body=search_body, search_pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME
-        )
+                # Search query.
+                query_text = "Python programming"
+                query_vector = _generate_test_vector(0.12)
+                search_body = DocumentQuery.get_hybrid_search_query(
+                    query_text=query_text,
+                    query_vector=query_vector,
+                    num_hits=5,
+                    tenant_state=tenant_state,
+                    # We're not worried about filtering here. tenant_id in this object
+                    # is not relevant.
+                    index_filters=IndexFilters(
+                        access_control_list=None, tenant_id=None
+                    ),
+                    include_hidden=False,
+                )
 
-        # Postcondition.
-        assert len(results) == len(docs)
-        # Assert that all the chunks above are present.
-        assert all(chunk.document_chunk.document_id in docs.keys() for chunk in results)
-        # Make sure the chunk contents are preserved.
-        for i, chunk in enumerate(results):
-            assert chunk.document_chunk == docs[chunk.document_chunk.document_id]
-            # Make sure score reporting seems reasonable (it should not be None
-            # or 0).
-            assert chunk.score
-            # Make sure there is some kind of match highlight only for the first
-            # result. The other results are so bad they're not expected to have
-            # match highlights.
-            if i == 0:
-                assert chunk.match_highlights.get(CONTENT_FIELD_NAME, [])
+                # Under test.
+                results = test_client.search(
+                    body=search_body, search_pipeline_id=pipeline_name
+                )
+
+                # Postcondition.
+                assert len(results) == len(docs)
+                # Assert that all the chunks above are present.
+                assert all(
+                    chunk.document_chunk.document_id in docs.keys() for chunk in results
+                )
+                # Make sure the chunk contents are preserved.
+                for i, chunk in enumerate(results):
+                    expected = docs[chunk.document_chunk.document_id]
+                    assert chunk.document_chunk == DocumentChunkWithoutVectors(
+                        **{
+                            k: getattr(expected, k)
+                            for k in DocumentChunkWithoutVectors.model_fields
+                        }
+                    )
+                    # Make sure score reporting seems reasonable (it should not be None
+                    # or 0).
+                    assert chunk.score
+                    # Make sure there is some kind of match highlight only for the first
+                    # result. The other results are so bad they're not expected to have
+                    # match highlights.
+                    if i == 0:
+                        assert chunk.match_highlights.get(CONTENT_FIELD_NAME, [])
 
     def test_search_empty_index(
         self,
@@ -828,7 +929,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
         # Note no documents were indexed.
 
@@ -845,11 +946,10 @@ class TestOpenSearchClient:
             index_filters=IndexFilters(access_control_list=None, tenant_id=None),
             include_hidden=False,
         )
+        pipeline_name, _ = get_normalization_pipeline_name_and_config()
 
         # Under test.
-        results = test_client.search(
-            body=search_body, search_pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME
-        )
+        results = test_client.search(body=search_body, search_pipeline_id=pipeline_name)
 
         # Postcondition.
         assert len(results) == 0
@@ -865,12 +965,13 @@ class TestOpenSearchClient:
         """
         # Precondition.
         _patch_global_tenant_state(monkeypatch, True)
+        _patch_opensearch_match_highlights_disabled(monkeypatch, False)
         tenant_x = TenantState(tenant_id="tenant-x", multitenant=True)
         tenant_y = TenantState(tenant_id="tenant-y", multitenant=True)
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_x.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Index documents with different public/hidden and tenant states.
@@ -948,11 +1049,10 @@ class TestOpenSearchClient:
             ),
             include_hidden=False,
         )
+        pipeline_name, _ = get_normalization_pipeline_name_and_config()
 
         # Under test.
-        results = test_client.search(
-            body=search_body, search_pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME
-        )
+        results = test_client.search(body=search_body, search_pipeline_id=pipeline_name)
 
         # Postcondition.
         # Should only get the public, non-hidden document, and the private
@@ -962,7 +1062,12 @@ class TestOpenSearchClient:
         # ordered; we're just assuming which doc will be the first result here.
         assert results[0].document_chunk.document_id == "public-doc"
         # Make sure the chunk contents are preserved.
-        assert results[0].document_chunk == docs["public-doc"]
+        assert results[0].document_chunk == DocumentChunkWithoutVectors(
+            **{
+                k: getattr(docs["public-doc"], k)
+                for k in DocumentChunkWithoutVectors.model_fields
+            }
+        )
         # Make sure score reporting seems reasonable (it should not be None
         # or 0).
         assert results[0].score
@@ -970,7 +1075,12 @@ class TestOpenSearchClient:
         assert results[0].match_highlights.get(CONTENT_FIELD_NAME, [])
         # Same for the second result.
         assert results[1].document_chunk.document_id == "private-doc-user-a"
-        assert results[1].document_chunk == docs["private-doc-user-a"]
+        assert results[1].document_chunk == DocumentChunkWithoutVectors(
+            **{
+                k: getattr(docs["private-doc-user-a"], k)
+                for k in DocumentChunkWithoutVectors.model_fields
+            }
+        )
         assert results[1].score
         assert results[1].match_highlights.get(CONTENT_FIELD_NAME, [])
 
@@ -986,11 +1096,12 @@ class TestOpenSearchClient:
         """
         # Precondition.
         _patch_global_tenant_state(monkeypatch, True)
+        _patch_opensearch_match_highlights_disabled(monkeypatch, False)
         tenant_x = TenantState(tenant_id="tenant-x", multitenant=True)
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_x.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Index documents with varying relevance to the query.
@@ -1067,11 +1178,10 @@ class TestOpenSearchClient:
             index_filters=IndexFilters(access_control_list=[], tenant_id=None),
             include_hidden=False,
         )
+        pipeline_name, _ = get_normalization_pipeline_name_and_config()
 
         # Under test.
-        results = test_client.search(
-            body=search_body, search_pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME
-        )
+        results = test_client.search(body=search_body, search_pipeline_id=pipeline_name)
 
         # Postcondition.
         # Should only get public, non-hidden documents (3 out of 5).
@@ -1118,7 +1228,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_x.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Although very unlikely in practice, let's use the same doc ID just to
@@ -1211,7 +1321,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Don't index any documents.
@@ -1238,7 +1348,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Index chunks for two different documents.
@@ -1306,7 +1416,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Index documents with different public/hidden and tenant states.
@@ -1383,7 +1493,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Index docs with various ages.
@@ -1441,15 +1551,16 @@ class TestOpenSearchClient:
             ),
             include_hidden=False,
         )
+        pipeline_name, _ = get_normalization_pipeline_name_and_config()
 
         # Under test.
         last_week_results = test_client.search(
             body=last_week_search_body,
-            search_pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME,
+            search_pipeline_id=pipeline_name,
         )
         last_six_months_results = test_client.search(
             body=last_six_months_search_body,
-            search_pipeline_id=MIN_MAX_NORMALIZATION_PIPELINE_NAME,
+            search_pipeline_id=pipeline_name,
         )
 
         # Postcondition.
@@ -1474,7 +1585,7 @@ class TestOpenSearchClient:
         mappings = DocumentSchema.get_document_schema(
             vector_dimension=128, multitenant=tenant_state.multitenant
         )
-        settings = DocumentSchema.get_index_settings()
+        settings = DocumentSchema.get_index_settings_based_on_environment()
         test_client.create_index(mappings=mappings, settings=settings)
 
         # Index chunks for two different documents, one hidden one not.
@@ -1523,4 +1634,281 @@ class TestOpenSearchClient:
         for result in results:
             # Note each result must be from doc 1, which is not hidden.
             expected_result = doc1_chunks[result.document_chunk.chunk_index]
-            assert result.document_chunk == expected_result
+            assert result.document_chunk == DocumentChunkWithoutVectors(
+                **{
+                    k: getattr(expected_result, k)
+                    for k in DocumentChunkWithoutVectors.model_fields
+                }
+            )
+
+    def test_keyword_search(
+        self,
+        test_client: OpenSearchIndexClient,
+        monkeypatch: pytest.MonkeyPatch,
+    ) -> None:
+        """
+        Tests keyword search with filters for ACL, hidden documents, and tenant
+        isolation.
+        """
+        # Precondition.
+        _patch_global_tenant_state(monkeypatch, True)
+        _patch_opensearch_match_highlights_disabled(monkeypatch, False)
+        tenant_x = TenantState(tenant_id="tenant-x", multitenant=True)
+        tenant_y = TenantState(tenant_id="tenant-y", multitenant=True)
+        mappings = DocumentSchema.get_document_schema(
+            vector_dimension=128, multitenant=tenant_x.multitenant
+        )
+        settings = DocumentSchema.get_index_settings_based_on_environment()
+        test_client.create_index(mappings=mappings, settings=settings)
+
+        # Index documents with different public/hidden and tenant states.
+        docs = {
+            "public-doc": _create_test_document_chunk(
+                document_id="public-doc",
+                chunk_index=0,
+                content="Public document content",
+                hidden=False,
+                tenant_state=tenant_x,
+            ),
+            "hidden-doc": _create_test_document_chunk(
+                document_id="hidden-doc",
+                chunk_index=0,
+                content="Hidden document content, spooky",
+                hidden=True,
+                tenant_state=tenant_x,
+            ),
+            "private-doc-user-a": _create_test_document_chunk(
+                document_id="private-doc-user-a",
+                chunk_index=0,
+                content="Private document content, btw my SSN is 123-45-6789",
+                hidden=False,
+                tenant_state=tenant_x,
+                document_access=DocumentAccess.build(
+                    user_emails=["user-a@example.com"],
+                    user_groups=[],
+                    external_user_emails=[],
+                    external_user_group_ids=[],
+                    is_public=False,
+                ),
+            ),
+            # Tests that we don't return documents that don't match keywords at
+            # all, even if they match filters.
+            "private-but-not-relevant-doc-user-a": _create_test_document_chunk(
+                document_id="private-but-not-relevant-doc-user-a",
+                chunk_index=0,
+                content="This text should not match the query at all",
+                hidden=False,
+                tenant_state=tenant_x,
+                document_access=DocumentAccess.build(
+                    user_emails=["user-a@example.com"],
+                    user_groups=[],
+                    external_user_emails=[],
+                    external_user_group_ids=[],
+                    is_public=False,
+                ),
+            ),
+            "private-doc-user-b": _create_test_document_chunk(
+                document_id="private-doc-user-b",
+                chunk_index=0,
+                content="Private document content, btw my SSN is 987-65-4321",
+                hidden=False,
+                tenant_state=tenant_x,
+                document_access=DocumentAccess.build(
+                    user_emails=["user-b@example.com"],
+                    user_groups=[],
+                    external_user_emails=[],
+                    external_user_group_ids=[],
+                    is_public=False,
+                ),
+            ),
+            "should-not-exist-from-tenant-x-pov": _create_test_document_chunk(
+                document_id="should-not-exist-from-tenant-x-pov",
+                chunk_index=0,
+                content="This is an entirely different tenant, x should never see this",
+                # Make this as permissive as possible to exercise tenant
+                # isolation.
+                hidden=False,
+                tenant_state=tenant_y,
+            ),
+        }
+        for doc in docs.values():
+            test_client.index_document(document=doc, tenant_state=doc.tenant_id)
+
+        # Refresh index to make documents searchable.
+        test_client.refresh_index()
+
+        # Should not match private-but-not-relevant-doc-user-a.
+        query_text = "document content"
+        search_body = DocumentQuery.get_keyword_search_query(
+            query_text=query_text,
+            num_hits=5,
+            tenant_state=tenant_x,
+            # The user should only be able to see their private docs. tenant_id
+            # in this object is not relevant.
+            index_filters=IndexFilters(
+                access_control_list=[prefix_user_email("user-a@example.com")],
+                tenant_id=None,
+            ),
+            include_hidden=False,
+        )
+
+        # Under test.
+        results = test_client.search(body=search_body, search_pipeline_id=None)
+
+        # Postcondition.
+        # Should only get the public, non-hidden document, and the private
+        # document for which the user has access.
+        assert len(results) == 2
+        # This should be the highest-ranked result, as a higher percentage of
+        # the content matches the query.
+        assert results[0].document_chunk.document_id == "public-doc"
+        # Make sure the chunk contents are preserved.
+        assert results[0].document_chunk == DocumentChunkWithoutVectors(
+            **{
+                k: getattr(docs["public-doc"], k)
+                for k in DocumentChunkWithoutVectors.model_fields
+            }
+        )
+        # Make sure score reporting seems reasonable (it should not be None
+        # or 0).
+        assert results[0].score
+        # Make sure there is some kind of match highlight.
+        assert results[0].match_highlights.get(CONTENT_FIELD_NAME, [])
+        # Same for the second result.
+        assert results[1].document_chunk.document_id == "private-doc-user-a"
+        assert results[1].document_chunk == DocumentChunkWithoutVectors(
+            **{
+                k: getattr(docs["private-doc-user-a"], k)
+                for k in DocumentChunkWithoutVectors.model_fields
+            }
+        )
+        assert results[1].score
+        assert results[1].match_highlights.get(CONTENT_FIELD_NAME, [])
+        assert results[1].score < results[0].score
+
+    def test_semantic_search(
+        self,
+        test_client: OpenSearchIndexClient,
+        monkeypatch: pytest.MonkeyPatch,
+    ) -> None:
+        """
+        Tests semantic search with filters for ACL, hidden documents, and tenant
+        isolation.
+        """
+        # Precondition.
+        _patch_global_tenant_state(monkeypatch, True)
+        tenant_x = TenantState(tenant_id="tenant-x", multitenant=True)
+        tenant_y = TenantState(tenant_id="tenant-y", multitenant=True)
+        mappings = DocumentSchema.get_document_schema(
+            vector_dimension=128, multitenant=tenant_x.multitenant
+        )
+        settings = DocumentSchema.get_index_settings_based_on_environment()
+        test_client.create_index(mappings=mappings, settings=settings)
+
+        # Index documents with different public/hidden and tenant states.
+        docs = {
+            "public-doc": _create_test_document_chunk(
+                document_id="public-doc",
+                chunk_index=0,
+                content="Public document content",
+                hidden=False,
+                tenant_state=tenant_x,
+                # Make this identical to the query vector to test that this
+                # result is returned first.
+                content_vector=_generate_test_vector(0.6),
+            ),
+            "hidden-doc": _create_test_document_chunk(
+                document_id="hidden-doc",
+                chunk_index=0,
+                content="Hidden document content, spooky",
+                hidden=True,
+                tenant_state=tenant_x,
+            ),
+            "private-doc-user-a": _create_test_document_chunk(
+                document_id="private-doc-user-a",
+                chunk_index=0,
+                content="Private document content, btw my SSN is 123-45-6789",
+                hidden=False,
+                tenant_state=tenant_x,
+                document_access=DocumentAccess.build(
+                    user_emails=["user-a@example.com"],
+                    user_groups=[],
+                    external_user_emails=[],
+                    external_user_group_ids=[],
+                    is_public=False,
+                ),
+                # Make this different from the query vector to test that this
+                # result is returned second.
+                content_vector=_generate_test_vector(0.5),
+            ),
+            "private-doc-user-b": _create_test_document_chunk(
+                document_id="private-doc-user-b",
+                chunk_index=0,
+                content="Private document content, btw my SSN is 987-65-4321",
+                hidden=False,
+                tenant_state=tenant_x,
+                document_access=DocumentAccess.build(
+                    user_emails=["user-b@example.com"],
+                    user_groups=[],
+                    external_user_emails=[],
+                    external_user_group_ids=[],
+                    is_public=False,
+                ),
+            ),
+            "should-not-exist-from-tenant-x-pov": _create_test_document_chunk(
+                document_id="should-not-exist-from-tenant-x-pov",
+                chunk_index=0,
+                content="This is an entirely different tenant, x should never see this",
+                # Make this as permissive as possible to exercise tenant
+                # isolation.
+                hidden=False,
+                tenant_state=tenant_y,
+            ),
+        }
+        for doc in docs.values():
+            test_client.index_document(document=doc, tenant_state=doc.tenant_id)
+
+        # Refresh index to make documents searchable.
+        test_client.refresh_index()
+
+        query_vector = _generate_test_vector(0.6)
+        search_body = DocumentQuery.get_semantic_search_query(
+            query_embedding=query_vector,
+            num_hits=5,
+            tenant_state=tenant_x,
+            # The user should only be able to see their private docs. tenant_id
+            # in this object is not relevant.
+            index_filters=IndexFilters(
+                access_control_list=[prefix_user_email("user-a@example.com")],
+                tenant_id=None,
+            ),
+            include_hidden=False,
+        )
+
+        # Under test.
+        results = test_client.search(body=search_body, search_pipeline_id=None)
+
+        # Postcondition.
+        # Should only get the public, non-hidden document, and the private
+        # document for which the user has access.
+        assert len(results) == 2
+        # We explicitly expect this to be the highest-ranked result.
+        assert results[0].document_chunk.document_id == "public-doc"
+        # Make sure the chunk contents are preserved.
+        assert results[0].document_chunk == DocumentChunkWithoutVectors(
+            **{
+                k: getattr(docs["public-doc"], k)
+                for k in DocumentChunkWithoutVectors.model_fields
+            }
+        )
+        assert results[0].score == 1.0
+        # Same for the second result.
+        assert results[1].document_chunk.document_id == "private-doc-user-a"
+        assert results[1].document_chunk == DocumentChunkWithoutVectors(
+            **{
+                k: getattr(docs["private-doc-user-a"], k)
+                for k in DocumentChunkWithoutVectors.model_fields
+            }
+        )
+        assert results[1].score
+        assert 0.0 < results[1].score < 1.0

backend/tests/external_dependency_unit/opensearch_migration/test_opensearch_migration_tasks.py

@@ -31,7 +31,6 @@ from onyx.background.celery.tasks.opensearch_migration.transformer import (
 )
 from onyx.configs.constants import PUBLIC_DOC_PAT
 from onyx.configs.constants import SOURCE_TYPE
-from onyx.context.search.models import IndexFilters
 from onyx.db.engine.sql_engine import get_session_with_current_tenant
 from onyx.db.models import Document
 from onyx.db.models import OpenSearchDocumentMigrationRecord
@@ -44,6 +43,7 @@ from onyx.document_index.opensearch.client import OpenSearchIndexClient
 from onyx.document_index.opensearch.client import wait_for_opensearch_with_timeout
 from onyx.document_index.opensearch.constants import DEFAULT_MAX_CHUNK_SIZE
 from onyx.document_index.opensearch.schema import DocumentChunk
+from onyx.document_index.opensearch.schema import get_opensearch_doc_chunk_id
 from onyx.document_index.opensearch.search import DocumentQuery
 from onyx.document_index.vespa.shared_utils.utils import wait_for_vespa_with_timeout
 from onyx.document_index.vespa.vespa_document_index import VespaDocumentIndex
@@ -70,6 +70,7 @@ from onyx.document_index.vespa_constants import SOURCE_LINKS
 from onyx.document_index.vespa_constants import TITLE
 from onyx.document_index.vespa_constants import TITLE_EMBEDDING
 from onyx.document_index.vespa_constants import USER_PROJECT
+from shared_configs.configs import MULTI_TENANT
 from shared_configs.contextvars import get_current_tenant_id
 from tests.external_dependency_unit.full_setup import ensure_full_deployment_setup
 
@@ -78,24 +79,22 @@ CHUNK_COUNT = 5
 
 
 def _get_document_chunks_from_opensearch(
-    opensearch_client: OpenSearchIndexClient, document_id: str, current_tenant_id: str
+    opensearch_client: OpenSearchIndexClient,
+    document_id: str,
+    tenant_state: TenantState,
 ) -> list[DocumentChunk]:
     opensearch_client.refresh_index()
-    filters = IndexFilters(access_control_list=None, tenant_id=current_tenant_id)
-    query_body = DocumentQuery.get_from_document_id_query(
-        document_id=document_id,
-        tenant_state=TenantState(tenant_id=current_tenant_id, multitenant=False),
-        index_filters=filters,
-        include_hidden=False,
-        max_chunk_size=DEFAULT_MAX_CHUNK_SIZE,
-        min_chunk_index=None,
-        max_chunk_index=None,
-    )
-    search_hits = opensearch_client.search(
-        body=query_body,
-        search_pipeline_id=None,
-    )
-    return [search_hit.document_chunk for search_hit in search_hits]
+    results: list[DocumentChunk] = []
+    for i in range(CHUNK_COUNT):
+        document_chunk_id: str = get_opensearch_doc_chunk_id(
+            tenant_state=tenant_state,
+            document_id=document_id,
+            chunk_index=i,
+            max_chunk_size=DEFAULT_MAX_CHUNK_SIZE,
+        )
+        result = opensearch_client.get_document(document_chunk_id)
+        results.append(result)
+    return results
 
 
 def _delete_document_chunks_from_opensearch(
@@ -452,10 +451,13 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         for chunks in document_chunks.values():
             all_chunks.extend(chunks)
         vespa_document_index.index_raw_chunks(all_chunks)
+        tenant_state = TenantState(
+            tenant_id=get_current_tenant_id(), multitenant=MULTI_TENANT
+        )
 
         # Under test.
         result = migrate_chunks_from_vespa_to_opensearch_task(
-            tenant_id=get_current_tenant_id()
+            tenant_id=tenant_state.tenant_id
         )
 
         # Postcondition.
@@ -477,7 +479,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         # Verify chunks were indexed in OpenSearch.
         for document in test_documents:
             opensearch_chunks = _get_document_chunks_from_opensearch(
-                opensearch_client, document.id, get_current_tenant_id()
+                opensearch_client, document.id, tenant_state
             )
             assert len(opensearch_chunks) == CHUNK_COUNT
             opensearch_chunks.sort(key=lambda x: x.chunk_index)
@@ -522,6 +524,9 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         for chunks in document_chunks.values():
             all_chunks.extend(chunks)
         vespa_document_index.index_raw_chunks(all_chunks)
+        tenant_state = TenantState(
+            tenant_id=get_current_tenant_id(), multitenant=MULTI_TENANT
+        )
 
         # Run the initial batch. To simulate partial progress we will mock the
         # redis lock to return True for the first invocation of .owned() and
@@ -536,7 +541,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
             return_value=mock_redis_client,
         ):
             result_1 = migrate_chunks_from_vespa_to_opensearch_task(
-                tenant_id=get_current_tenant_id()
+                tenant_id=tenant_state.tenant_id
             )
 
         assert result_1 is True
@@ -559,7 +564,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         # Under test.
         # Run the remainder of the migration.
         result_2 = migrate_chunks_from_vespa_to_opensearch_task(
-            tenant_id=get_current_tenant_id()
+            tenant_id=tenant_state.tenant_id
         )
 
         # Postcondition.
@@ -583,7 +588,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         # Verify chunks were indexed in OpenSearch.
         for document in test_documents:
             opensearch_chunks = _get_document_chunks_from_opensearch(
-                opensearch_client, document.id, get_current_tenant_id()
+                opensearch_client, document.id, tenant_state
             )
             assert len(opensearch_chunks) == CHUNK_COUNT
             opensearch_chunks.sort(key=lambda x: x.chunk_index)
@@ -630,6 +635,9 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         for chunks in document_chunks.values():
             all_chunks.extend(chunks)
         vespa_document_index.index_raw_chunks(all_chunks)
+        tenant_state = TenantState(
+            tenant_id=get_current_tenant_id(), multitenant=MULTI_TENANT
+        )
 
         # Run the initial batch. To simulate partial progress we will mock the
         # redis lock to return True for the first invocation of .owned() and
@@ -646,7 +654,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
             return_value=mock_redis_client,
         ):
             result_1 = migrate_chunks_from_vespa_to_opensearch_task(
-                tenant_id=get_current_tenant_id()
+                tenant_id=tenant_state.tenant_id
             )
 
         assert result_1 is True
@@ -691,7 +699,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
             ),
         ):
             result_2 = migrate_chunks_from_vespa_to_opensearch_task(
-                tenant_id=get_current_tenant_id()
+                tenant_id=tenant_state.tenant_id
             )
 
         # Postcondition.
@@ -728,7 +736,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
             ),
         ):
             result_3 = migrate_chunks_from_vespa_to_opensearch_task(
-                tenant_id=get_current_tenant_id()
+                tenant_id=tenant_state.tenant_id
             )
 
         # Postcondition.
@@ -752,7 +760,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         # Verify chunks were indexed in OpenSearch.
         for document in test_documents:
             opensearch_chunks = _get_document_chunks_from_opensearch(
-                opensearch_client, document.id, get_current_tenant_id()
+                opensearch_client, document.id, tenant_state
             )
             assert len(opensearch_chunks) == CHUNK_COUNT
             opensearch_chunks.sort(key=lambda x: x.chunk_index)
@@ -840,24 +848,25 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
             chunk["content"] = (
                 f"Different content {chunk[CHUNK_ID]} for {test_documents[0].id}"
             )
+        tenant_state = TenantState(
+            tenant_id=get_current_tenant_id(), multitenant=MULTI_TENANT
+        )
         chunks_for_document_in_opensearch, _ = (
             transform_vespa_chunks_to_opensearch_chunks(
                 document_in_opensearch,
-                TenantState(tenant_id=get_current_tenant_id(), multitenant=False),
+                tenant_state,
                 {},
             )
         )
         opensearch_client.bulk_index_documents(
             documents=chunks_for_document_in_opensearch,
-            tenant_state=TenantState(
-                tenant_id=get_current_tenant_id(), multitenant=False
-            ),
+            tenant_state=tenant_state,
             update_if_exists=True,
         )
 
         # Under test.
         result = migrate_chunks_from_vespa_to_opensearch_task(
-            tenant_id=get_current_tenant_id()
+            tenant_id=tenant_state.tenant_id
         )
 
         # Postcondition.
@@ -878,7 +887,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         # Verify chunks were indexed in OpenSearch.
         for document in test_documents:
             opensearch_chunks = _get_document_chunks_from_opensearch(
-                opensearch_client, document.id, get_current_tenant_id()
+                opensearch_client, document.id, tenant_state
             )
             assert len(opensearch_chunks) == CHUNK_COUNT
             opensearch_chunks.sort(key=lambda x: x.chunk_index)
@@ -922,11 +931,14 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         for chunks in document_chunks.values():
             all_chunks.extend(chunks)
         vespa_document_index.index_raw_chunks(all_chunks)
+        tenant_state = TenantState(
+            tenant_id=get_current_tenant_id(), multitenant=MULTI_TENANT
+        )
 
         # Under test.
         # First run.
         result_1 = migrate_chunks_from_vespa_to_opensearch_task(
-            tenant_id=get_current_tenant_id()
+            tenant_id=tenant_state.tenant_id
         )
 
         # Postcondition.
@@ -947,7 +959,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         # Verify chunks were indexed in OpenSearch.
         for document in test_documents:
             opensearch_chunks = _get_document_chunks_from_opensearch(
-                opensearch_client, document.id, get_current_tenant_id()
+                opensearch_client, document.id, tenant_state
             )
             assert len(opensearch_chunks) == CHUNK_COUNT
             opensearch_chunks.sort(key=lambda x: x.chunk_index)
@@ -960,7 +972,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         # Under test.
         # Second run.
         result_2 = migrate_chunks_from_vespa_to_opensearch_task(
-            tenant_id=get_current_tenant_id()
+            tenant_id=tenant_state.tenant_id
         )
 
         # Postcondition.
@@ -982,7 +994,7 @@ class TestMigrateChunksFromVespaToOpenSearchTask:
         # Verify chunks were indexed in OpenSearch.
         for document in test_documents:
             opensearch_chunks = _get_document_chunks_from_opensearch(
-                opensearch_client, document.id, get_current_tenant_id()
+                opensearch_client, document.id, tenant_state
             )
             assert len(opensearch_chunks) == CHUNK_COUNT
             opensearch_chunks.sort(key=lambda x: x.chunk_index)

backend/tests/external_dependency_unit/tools/test_python_tool.py

@@ -1219,15 +1219,16 @@ def test_code_interpreter_receives_chat_files(
         finally:
             ci_mod.CodeInterpreterClient.__init__.__defaults__ = original_defaults
 
-    # Verify: file uploaded, code executed via streaming, staged file cleaned up
+    # Verify: file uploaded and code executed via streaming.
     assert len(mock_ci_server.get_requests(method="POST", path="/v1/files")) == 1
     assert (
         len(mock_ci_server.get_requests(method="POST", path="/v1/execute/stream")) == 1
     )
 
-    delete_requests = mock_ci_server.get_requests(method="DELETE")
-    assert len(delete_requests) == 1
-    assert delete_requests[0].path.startswith("/v1/files/")
+    # Staged input files are intentionally NOT deleted — PythonTool caches their
+    # file IDs across agent-loop iterations to avoid re-uploading on every call.
+    # The code interpreter cleans them up via its own TTL.
+    assert len(mock_ci_server.get_requests(method="DELETE")) == 0
 
     execute_body = mock_ci_server.get_requests(
         method="POST", path="/v1/execute/stream"

backend/tests/unit/onyx/chat/test_llm_loop.py

@@ -1,9 +1,13 @@
-"""Tests for llm_loop.py, specifically the construct_message_history function."""
+"""Tests for llm_loop.py, including history construction and empty-response paths."""
+
+from unittest.mock import Mock
 
 import pytest
 
+from onyx.chat.llm_loop import _build_empty_llm_response_error
 from onyx.chat.llm_loop import _try_fallback_tool_extraction
 from onyx.chat.llm_loop import construct_message_history
+from onyx.chat.llm_loop import EmptyLLMResponseError
 from onyx.chat.models import ChatLoadedFile
 from onyx.chat.models import ChatMessageSimple
 from onyx.chat.models import ContextFileMetadata
@@ -13,6 +17,7 @@ from onyx.chat.models import LlmStepResult
 from onyx.chat.models import ToolCallSimple
 from onyx.configs.constants import MessageType
 from onyx.file_store.models import ChatFileType
+from onyx.llm.interfaces import LLMConfig
 from onyx.llm.interfaces import ToolChoiceOptions
 from onyx.server.query_and_chat.placement import Placement
 from onyx.tools.models import ToolCallKickoff
@@ -1167,3 +1172,57 @@ class TestFallbackToolExtraction:
 
         assert result is llm_step_result
         assert attempted is False
+
+
+class TestEmptyLlmResponseClassification:
+    def _make_llm(self, provider: str = "openai", model: str = "gpt-5.2") -> Mock:
+        llm = Mock()
+        llm.config = LLMConfig(
+            model_provider=provider,
+            model_name=model,
+            temperature=0.0,
+            max_input_tokens=4096,
+        )
+        return llm
+
+    def test_openai_empty_stream_is_classified_as_budget_exceeded(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        monkeypatch.setattr("onyx.chat.llm_loop.is_true_openai_model", lambda *_: True)
+
+        err = _build_empty_llm_response_error(
+            llm=self._make_llm(),
+            llm_step_result=LlmStepResult(
+                reasoning=None,
+                answer=None,
+                tool_calls=None,
+                raw_answer=None,
+            ),
+            tool_choice=ToolChoiceOptions.AUTO,
+        )
+
+        assert isinstance(err, EmptyLLMResponseError)
+        assert err.error_code == "BUDGET_EXCEEDED"
+        assert err.is_retryable is False
+        assert "quota" in err.client_error_msg.lower()
+
+    def test_reasoning_only_response_uses_generic_empty_response_error(
+        self, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        monkeypatch.setattr("onyx.chat.llm_loop.is_true_openai_model", lambda *_: True)
+
+        err = _build_empty_llm_response_error(
+            llm=self._make_llm(),
+            llm_step_result=LlmStepResult(
+                reasoning="scratchpad only",
+                answer=None,
+                tool_calls=None,
+                raw_answer=None,
+            ),
+            tool_choice=ToolChoiceOptions.AUTO,
+        )
+
+        assert isinstance(err, EmptyLLMResponseError)
+        assert err.error_code == "EMPTY_LLM_RESPONSE"
+        assert err.is_retryable is True
+        assert "quota" not in err.client_error_msg.lower()

backend/tests/unit/onyx/chat/test_process_message.py

@@ -0,0 +1,133 @@
+import pytest
+
+from onyx.chat.process_message import _resolve_query_processing_hook_result
+from onyx.chat.process_message import remove_answer_citations
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
+from onyx.hooks.executor import HookSkipped
+from onyx.hooks.executor import HookSoftFailed
+from onyx.hooks.points.query_processing import QueryProcessingResponse
+
+
+def test_remove_answer_citations_strips_http_markdown_citation() -> None:
+    answer = "The answer is Paris [[1]](https://example.com/doc)."
+
+    assert remove_answer_citations(answer) == "The answer is Paris."
+
+
+def test_remove_answer_citations_strips_empty_markdown_citation() -> None:
+    answer = "The answer is Paris [[1]]()."
+
+    assert remove_answer_citations(answer) == "The answer is Paris."
+
+
+def test_remove_answer_citations_strips_citation_with_parentheses_in_url() -> None:
+    answer = (
+        "The answer is Paris "
+        "[[1]](https://en.wikipedia.org/wiki/Function_(mathematics))."
+    )
+
+    assert remove_answer_citations(answer) == "The answer is Paris."
+
+
+def test_remove_answer_citations_preserves_non_citation_markdown_links() -> None:
+    answer = (
+        "See [reference](https://example.com/Function_(mathematics)) "
+        "for context [[1]](https://en.wikipedia.org/wiki/Function_(mathematics))."
+    )
+
+    assert (
+        remove_answer_citations(answer)
+        == "See [reference](https://example.com/Function_(mathematics)) for context."
+    )
+
+
+# ---------------------------------------------------------------------------
+# Query Processing hook response handling (_resolve_query_processing_hook_result)
+# ---------------------------------------------------------------------------
+
+
+def test_wrong_model_type_raises_internal_error() -> None:
+    """If the executor ever returns an unexpected BaseModel type, raise INTERNAL_ERROR
+    rather than an AssertionError or AttributeError."""
+    from pydantic import BaseModel as PydanticBaseModel
+
+    class _OtherModel(PydanticBaseModel):
+        pass
+
+    with pytest.raises(OnyxError) as exc_info:
+        _resolve_query_processing_hook_result(_OtherModel(), "original query")
+    assert exc_info.value.error_code is OnyxErrorCode.INTERNAL_ERROR
+
+
+def test_hook_skipped_leaves_message_text_unchanged() -> None:
+    result = _resolve_query_processing_hook_result(HookSkipped(), "original query")
+    assert result == "original query"
+
+
+def test_hook_soft_failed_leaves_message_text_unchanged() -> None:
+    result = _resolve_query_processing_hook_result(HookSoftFailed(), "original query")
+    assert result == "original query"
+
+
+def test_null_query_raises_query_rejected() -> None:
+    with pytest.raises(OnyxError) as exc_info:
+        _resolve_query_processing_hook_result(
+            QueryProcessingResponse(query=None), "original query"
+        )
+    assert exc_info.value.error_code is OnyxErrorCode.QUERY_REJECTED
+
+
+def test_empty_string_query_raises_query_rejected() -> None:
+    """Empty string is falsy — must be treated as rejection, same as None."""
+    with pytest.raises(OnyxError) as exc_info:
+        _resolve_query_processing_hook_result(
+            QueryProcessingResponse(query=""), "original query"
+        )
+    assert exc_info.value.error_code is OnyxErrorCode.QUERY_REJECTED
+
+
+def test_whitespace_only_query_raises_query_rejected() -> None:
+    """Whitespace-only string is truthy but meaningless — must be treated as rejection."""
+    with pytest.raises(OnyxError) as exc_info:
+        _resolve_query_processing_hook_result(
+            QueryProcessingResponse(query="   "), "original query"
+        )
+    assert exc_info.value.error_code is OnyxErrorCode.QUERY_REJECTED
+
+
+def test_absent_query_field_raises_query_rejected() -> None:
+    """query defaults to None when not provided."""
+    with pytest.raises(OnyxError) as exc_info:
+        _resolve_query_processing_hook_result(
+            QueryProcessingResponse(), "original query"
+        )
+    assert exc_info.value.error_code is OnyxErrorCode.QUERY_REJECTED
+
+
+def test_rejection_message_surfaced_in_error_when_provided() -> None:
+    with pytest.raises(OnyxError) as exc_info:
+        _resolve_query_processing_hook_result(
+            QueryProcessingResponse(
+                query=None, rejection_message="Queries about X are not allowed."
+            ),
+            "original query",
+        )
+    assert "Queries about X are not allowed." in str(exc_info.value)
+
+
+def test_fallback_rejection_message_when_none() -> None:
+    """No rejection_message → generic fallback used in OnyxError detail."""
+    with pytest.raises(OnyxError) as exc_info:
+        _resolve_query_processing_hook_result(
+            QueryProcessingResponse(query=None, rejection_message=None),
+            "original query",
+        )
+    assert "Your query was rejected." in str(exc_info.value)
+
+
+def test_nonempty_query_rewrites_message_text() -> None:
+    result = _resolve_query_processing_hook_result(
+        QueryProcessingResponse(query="rewritten query"), "original query"
+    )
+    assert result == "rewritten query"

backend/tests/unit/onyx/chat/test_process_message_mock_llm.py

@@ -3,7 +3,10 @@ from unittest.mock import Mock
 import pytest
 
 from onyx.chat import process_message
+from onyx.chat.models import AnswerStream
+from onyx.chat.models import StreamingError
 from onyx.configs import app_configs
+from onyx.server.query_and_chat.models import MessageResponseIDInfo
 from onyx.server.query_and_chat.models import SendMessageRequest
 
 
@@ -35,3 +38,26 @@ def test_mock_llm_response_requires_integration_mode() -> None:
                 db_session=Mock(),
             )
         )
+
+
+def test_gather_stream_returns_empty_answer_when_streaming_error_only() -> None:
+    packets: AnswerStream = iter(
+        [
+            MessageResponseIDInfo(
+                user_message_id=None,
+                reserved_assistant_message_id=42,
+            ),
+            StreamingError(
+                error="OpenAI quota exceeded",
+                error_code="BUDGET_EXCEEDED",
+                is_retryable=False,
+            ),
+        ]
+    )
+
+    result = process_message.gather_stream(packets)
+
+    assert result.answer == ""
+    assert result.answer_citationless == ""
+    assert result.error_msg == "OpenAI quota exceeded"
+    assert result.message_id == 42

backend/tests/unit/onyx/db/test_projects_upload_task_expiry.py

@@ -0,0 +1,63 @@
+"""
+Unit test verifying that the upload API path sends tasks with expires=.
+
+The upload_files_to_user_files_with_indexing function must include expires=
+on every send_task call to prevent phantom task accumulation if the worker
+is down or slow.
+"""
+
+from unittest.mock import MagicMock
+from unittest.mock import patch
+from uuid import uuid4
+
+from onyx.configs.constants import CELERY_USER_FILE_PROCESSING_TASK_EXPIRES
+from onyx.configs.constants import OnyxCeleryQueues
+from onyx.configs.constants import OnyxCeleryTask
+from onyx.db.models import UserFile
+from onyx.db.projects import upload_files_to_user_files_with_indexing
+
+
+def _make_mock_user_file() -> MagicMock:
+    uf = MagicMock(spec=UserFile)
+    uf.id = str(uuid4())
+    return uf
+
+
+@patch("onyx.db.projects.get_current_tenant_id", return_value="test_tenant")
+@patch("onyx.db.projects.create_user_files")
+@patch(
+    "onyx.background.celery.versioned_apps.client.app",
+    new_callable=MagicMock,
+)
+def test_send_task_includes_expires(
+    mock_client_app: MagicMock,
+    mock_create: MagicMock,
+    mock_tenant: MagicMock,  # noqa: ARG001
+) -> None:
+    """Every send_task call from the upload path must include expires=."""
+    user_files = [_make_mock_user_file(), _make_mock_user_file()]
+    mock_create.return_value = MagicMock(
+        user_files=user_files,
+        rejected_files=[],
+        id_to_temp_id={},
+    )
+
+    mock_user = MagicMock()
+    mock_db_session = MagicMock()
+
+    upload_files_to_user_files_with_indexing(
+        files=[],
+        project_id=None,
+        user=mock_user,
+        temp_id_map=None,
+        db_session=mock_db_session,
+    )
+
+    assert mock_client_app.send_task.call_count == len(user_files)
+
+    for call in mock_client_app.send_task.call_args_list:
+        assert call.args[0] == OnyxCeleryTask.PROCESS_SINGLE_USER_FILE
+        assert call.kwargs.get("queue") == OnyxCeleryQueues.USER_FILE_PROCESSING
+        assert (
+            call.kwargs.get("expires") == CELERY_USER_FILE_PROCESSING_TASK_EXPIRES
+        ), "send_task must include expires= to prevent phantom task accumulation"

backend/tests/unit/onyx/file_processing/fixtures/empty.pdf

@@ -0,0 +1,45 @@
+%PDF-1.3
+%<25><><EFBFBD><EFBFBD>
+1 0 obj
+<<
+/Producer (pypdf)
+>>
+endobj
+2 0 obj
+<<
+/Type /Pages
+/Count 1
+/Kids [ 4 0 R ]
+>>
+endobj
+3 0 obj
+<<
+/Type /Catalog
+/Pages 2 0 R
+>>
+endobj
+4 0 obj
+<<
+/Type /Page
+/Resources <<
+>>
+/MediaBox [ 0.0 0.0 200 200 ]
+/Parent 2 0 R
+>>
+endobj
+xref
+0 5
+0000000000 65535 f 
+0000000015 00000 n 
+0000000054 00000 n 
+0000000113 00000 n 
+0000000162 00000 n 
+trailer
+<<
+/Size 5
+/Root 3 0 R
+/Info 1 0 R
+>>
+startxref
+256
+%%EOF

backend/tests/unit/onyx/file_processing/fixtures/multipage.pdf

@@ -0,0 +1,89 @@
+%PDF-1.3
+%<25><><EFBFBD><EFBFBD>
+1 0 obj
+<<
+/Producer (pypdf)
+>>
+endobj
+2 0 obj
+<<
+/Type /Pages
+/Count 2
+/Kids [ 4 0 R 6 0 R ]
+>>
+endobj
+3 0 obj
+<<
+/Type /Catalog
+/Pages 2 0 R
+>>
+endobj
+4 0 obj
+<<
+/Type /Page
+/Resources <<
+/Font <<
+/F1 <<
+/Type /Font
+/Subtype /Type1
+/BaseFont /Helvetica
+>>
+>>
+>>
+/MediaBox [ 0.0 0.0 200 200 ]
+/Contents 5 0 R
+/Parent 2 0 R
+>>
+endobj
+5 0 obj
+<<
+/Length 47
+>>
+stream
+BT /F1 12 Tf 50 150 Td (Page one content) Tj ET
+endstream
+endobj
+6 0 obj
+<<
+/Type /Page
+/Resources <<
+/Font <<
+/F1 <<
+/Type /Font
+/Subtype /Type1
+/BaseFont /Helvetica
+>>
+>>
+>>
+/MediaBox [ 0.0 0.0 200 200 ]
+/Contents 7 0 R
+/Parent 2 0 R
+>>
+endobj
+7 0 obj
+<<
+/Length 47
+>>
+stream
+BT /F1 12 Tf 50 150 Td (Page two content) Tj ET
+endstream
+endobj
+xref
+0 8
+0000000000 65535 f 
+0000000015 00000 n 
+0000000054 00000 n 
+0000000119 00000 n 
+0000000168 00000 n 
+0000000349 00000 n 
+0000000446 00000 n 
+0000000627 00000 n 
+trailer
+<<
+/Size 8
+/Root 3 0 R
+/Info 1 0 R
+>>
+startxref
+724
+%%EOF

backend/tests/unit/onyx/file_processing/fixtures/simple.pdf

@@ -0,0 +1,62 @@
+%PDF-1.3
+%<25><><EFBFBD><EFBFBD>
+1 0 obj
+<<
+/Producer (pypdf)
+>>
+endobj
+2 0 obj
+<<
+/Type /Pages
+/Count 1
+/Kids [ 4 0 R ]
+>>
+endobj
+3 0 obj
+<<
+/Type /Catalog
+/Pages 2 0 R
+>>
+endobj
+4 0 obj
+<<
+/Type /Page
+/Resources <<
+/Font <<
+/F1 <<
+/Type /Font
+/Subtype /Type1
+/BaseFont /Helvetica
+>>
+>>
+>>
+/MediaBox [ 0.0 0.0 200 200 ]
+/Contents 5 0 R
+/Parent 2 0 R
+>>
+endobj
+5 0 obj
+<<
+/Length 42
+>>
+stream
+BT /F1 12 Tf 50 150 Td (Hello World) Tj ET
+endstream
+endobj
+xref
+0 6
+0000000000 65535 f 
+0000000015 00000 n 
+0000000054 00000 n 
+0000000113 00000 n 
+0000000162 00000 n 
+0000000343 00000 n 
+trailer
+<<
+/Size 6
+/Root 3 0 R
+/Info 1 0 R
+>>
+startxref
+435
+%%EOF

backend/tests/unit/onyx/file_processing/fixtures/with_metadata.pdf

@@ -0,0 +1,64 @@
+%PDF-1.3
+%<25><><EFBFBD><EFBFBD>
+1 0 obj
+<<
+/Producer (pypdf)
+/Title (My Title)
+/Author (Jane Doe)
+>>
+endobj
+2 0 obj
+<<
+/Type /Pages
+/Count 1
+/Kids [ 4 0 R ]
+>>
+endobj
+3 0 obj
+<<
+/Type /Catalog
+/Pages 2 0 R
+>>
+endobj
+4 0 obj
+<<
+/Type /Page
+/Resources <<
+/Font <<
+/F1 <<
+/Type /Font
+/Subtype /Type1
+/BaseFont /Helvetica
+>>
+>>
+>>
+/MediaBox [ 0.0 0.0 200 200 ]
+/Contents 5 0 R
+/Parent 2 0 R
+>>
+endobj
+5 0 obj
+<<
+/Length 35
+>>
+stream
+BT /F1 12 Tf 50 150 Td (test) Tj ET
+endstream
+endobj
+xref
+0 6
+0000000000 65535 f 
+0000000015 00000 n 
+0000000091 00000 n 
+0000000150 00000 n 
+0000000199 00000 n 
+0000000380 00000 n 
+trailer
+<<
+/Size 6
+/Root 3 0 R
+/Info 1 0 R
+>>
+startxref
+465
+%%EOF

backend/tests/unit/onyx/file_processing/test_image_summarization_errors.py

@@ -0,0 +1,89 @@
+"""
+Unit tests for image summarization error handling.
+
+Verifies that:
+1. LLM errors produce actionable error messages (not base64 dumps)
+2. Unsupported MIME type logs include the magic bytes and size
+3. The ValueError raised on LLM failure preserves the original exception
+"""
+
+from unittest.mock import MagicMock
+from unittest.mock import patch
+
+import pytest
+
+from onyx.file_processing.image_summarization import _summarize_image
+from onyx.file_processing.image_summarization import summarize_image_with_error_handling
+
+
+class TestSummarizeImageErrorMessage:
+    """_summarize_image must not dump base64 image data into error messages."""
+
+    def test_error_message_contains_exception_type_not_base64(self) -> None:
+        """The ValueError should contain the original exception info, not message payloads."""
+        mock_llm = MagicMock()
+        mock_llm.invoke.side_effect = RuntimeError("Connection timeout")
+
+        # A fake base64-encoded image string (should NOT appear in the error)
+        fake_encoded = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUg..."
+
+        with pytest.raises(ValueError, match="RuntimeError: Connection timeout"):
+            _summarize_image(fake_encoded, mock_llm, query="test")
+
+    def test_error_message_does_not_contain_base64(self) -> None:
+        """Ensure base64 data is never included in the error message."""
+        mock_llm = MagicMock()
+        mock_llm.invoke.side_effect = RuntimeError("API error")
+
+        fake_encoded = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAUA"
+
+        with pytest.raises(ValueError) as exc_info:
+            _summarize_image(fake_encoded, mock_llm)
+
+        error_str = str(exc_info.value)
+        assert "base64" not in error_str
+        assert "iVBOR" not in error_str
+
+    def test_original_exception_is_chained(self) -> None:
+        """The ValueError should chain the original exception via __cause__."""
+        mock_llm = MagicMock()
+        original = RuntimeError("upstream failure")
+        mock_llm.invoke.side_effect = original
+
+        with pytest.raises(ValueError) as exc_info:
+            _summarize_image("data:image/png;base64,abc", mock_llm)
+
+        assert exc_info.value.__cause__ is original
+
+
+class TestUnsupportedMimeTypeLogging:
+    """summarize_image_with_error_handling should log useful info for unsupported formats."""
+
+    @patch(
+        "onyx.file_processing.image_summarization.summarize_image_pipeline",
+        side_effect=__import__(
+            "onyx.file_processing.image_summarization",
+            fromlist=["UnsupportedImageFormatError"],
+        ).UnsupportedImageFormatError("unsupported"),
+    )
+    def test_logs_magic_bytes_and_size(
+        self, mock_pipeline: MagicMock  # noqa: ARG002
+    ) -> None:
+        """The info log should include magic bytes hex and image size."""
+        mock_llm = MagicMock()
+        # TIFF magic bytes (not in the supported list)
+        image_data = b"\x49\x49\x2a\x00" + b"\x00" * 100
+
+        with patch("onyx.file_processing.image_summarization.logger") as mock_logger:
+            result = summarize_image_with_error_handling(
+                llm=mock_llm,
+                image_data=image_data,
+                context_name="test_image.tiff",
+            )
+
+        assert result is None
+        mock_logger.info.assert_called_once()
+        log_args = mock_logger.info.call_args
+        # Check the format string args contain magic bytes and size
+        assert "49492a00" in str(log_args)
+        assert "104" in str(log_args)  # 4 + 100 bytes

backend/tests/unit/onyx/file_processing/test_image_summarization_litellm_errors.py

@@ -0,0 +1,141 @@
+"""
+Unit tests verifying that LiteLLM error details are extracted and surfaced
+in image summarization error messages.
+
+When the LLM call fails, the error handler should include the status_code,
+llm_provider, and model from LiteLLM exceptions so operators can diagnose
+the root cause (rate limit, content filter, unsupported vision, etc.)
+without needing to dig through LiteLLM internals.
+"""
+
+from unittest.mock import MagicMock
+
+import pytest
+
+from onyx.file_processing.image_summarization import _summarize_image
+
+
+def _make_litellm_style_error(
+    *,
+    message: str = "API error",
+    status_code: int | None = None,
+    llm_provider: str | None = None,
+    model: str | None = None,
+) -> RuntimeError:
+    """Create an exception with LiteLLM-style attributes."""
+    exc = RuntimeError(message)
+    if status_code is not None:
+        exc.status_code = status_code  # type: ignore[attr-defined]
+    if llm_provider is not None:
+        exc.llm_provider = llm_provider  # type: ignore[attr-defined]
+    if model is not None:
+        exc.model = model  # type: ignore[attr-defined]
+    return exc
+
+
+class TestLiteLLMErrorExtraction:
+    """Verify that LiteLLM error attributes are included in the ValueError."""
+
+    def test_status_code_included(self) -> None:
+        mock_llm = MagicMock()
+        mock_llm.invoke.side_effect = _make_litellm_style_error(
+            message="Content filter triggered",
+            status_code=400,
+            llm_provider="azure",
+            model="gpt-4o",
+        )
+
+        with pytest.raises(ValueError, match="status_code=400"):
+            _summarize_image("data:image/png;base64,abc", mock_llm)
+
+    def test_llm_provider_included(self) -> None:
+        mock_llm = MagicMock()
+        mock_llm.invoke.side_effect = _make_litellm_style_error(
+            message="Bad request",
+            status_code=400,
+            llm_provider="azure",
+        )
+
+        with pytest.raises(ValueError, match="llm_provider=azure"):
+            _summarize_image("data:image/png;base64,abc", mock_llm)
+
+    def test_model_included(self) -> None:
+        mock_llm = MagicMock()
+        mock_llm.invoke.side_effect = _make_litellm_style_error(
+            message="Bad request",
+            model="gpt-4o",
+        )
+
+        with pytest.raises(ValueError, match="model=gpt-4o"):
+            _summarize_image("data:image/png;base64,abc", mock_llm)
+
+    def test_all_fields_in_single_message(self) -> None:
+        mock_llm = MagicMock()
+        mock_llm.invoke.side_effect = _make_litellm_style_error(
+            message="Rate limit exceeded",
+            status_code=429,
+            llm_provider="azure",
+            model="gpt-4o",
+        )
+
+        with pytest.raises(ValueError) as exc_info:
+            _summarize_image("data:image/png;base64,abc", mock_llm)
+
+        msg = str(exc_info.value)
+        assert "status_code=429" in msg
+        assert "llm_provider=azure" in msg
+        assert "model=gpt-4o" in msg
+        assert "Rate limit exceeded" in msg
+
+    def test_plain_exception_without_litellm_attrs(self) -> None:
+        """Non-LiteLLM exceptions should still produce a useful message."""
+        mock_llm = MagicMock()
+        mock_llm.invoke.side_effect = ConnectionError("Connection refused")
+
+        with pytest.raises(ValueError) as exc_info:
+            _summarize_image("data:image/png;base64,abc", mock_llm)
+
+        msg = str(exc_info.value)
+        assert "ConnectionError" in msg
+        assert "Connection refused" in msg
+        # Should not contain status_code/llm_provider/model
+        assert "status_code" not in msg
+        assert "llm_provider" not in msg
+
+    def test_no_base64_in_error(self) -> None:
+        """Error messages must not contain the full base64 image payload.
+
+        Some LiteLLM exceptions echo the request body (including base64 images)
+        in their message.  The truncation guard ensures the bulk of such a
+        payload is stripped from the re-raised ValueError.
+        """
+        mock_llm = MagicMock()
+        # Build a long base64-like payload that exceeds the 512-char truncation
+        fake_b64_payload = "iVBORw0KGgo" * 100  # ~1100 chars
+        fake_b64 = f"data:image/png;base64,{fake_b64_payload}"
+
+        mock_llm.invoke.side_effect = RuntimeError(
+            f"Request failed for payload: {fake_b64}"
+        )
+
+        with pytest.raises(ValueError) as exc_info:
+            _summarize_image(fake_b64, mock_llm)
+
+        msg = str(exc_info.value)
+        # The full payload must not appear (truncation should have kicked in)
+        assert fake_b64_payload not in msg
+        assert "truncated" in msg
+
+    def test_long_error_message_truncated(self) -> None:
+        """Exception messages longer than 512 chars are truncated."""
+        mock_llm = MagicMock()
+        long_msg = "x" * 1000
+        mock_llm.invoke.side_effect = RuntimeError(long_msg)
+
+        with pytest.raises(ValueError) as exc_info:
+            _summarize_image("data:image/png;base64,abc", mock_llm)
+
+        msg = str(exc_info.value)
+        assert "truncated" in msg
+        # The full 1000-char string should not appear
+        assert long_msg not in msg

backend/tests/unit/onyx/file_processing/test_pdf.py

@@ -0,0 +1,124 @@
+"""Unit tests for pypdf-dependent PDF processing functions.
+
+Tests cover:
+- read_pdf_file: text extraction, metadata, encrypted PDFs, image extraction
+- pdf_to_text: convenience wrapper
+- is_pdf_protected: password protection detection
+
+Fixture PDFs live in ./fixtures/ and are pre-built so the test layer has no
+dependency on pypdf internals (pypdf.generic).
+"""
+
+from io import BytesIO
+from pathlib import Path
+
+from onyx.file_processing.extract_file_text import pdf_to_text
+from onyx.file_processing.extract_file_text import read_pdf_file
+from onyx.file_processing.password_validation import is_pdf_protected
+
+FIXTURES = Path(__file__).parent / "fixtures"
+
+
+def _load(name: str) -> BytesIO:
+    return BytesIO((FIXTURES / name).read_bytes())
+
+
+# ── read_pdf_file ────────────────────────────────────────────────────────
+
+
+class TestReadPdfFile:
+    def test_basic_text_extraction(self) -> None:
+        text, _, images = read_pdf_file(_load("simple.pdf"))
+        assert "Hello World" in text
+        assert images == []
+
+    def test_multi_page_text_extraction(self) -> None:
+        text, _, _ = read_pdf_file(_load("multipage.pdf"))
+        assert "Page one content" in text
+        assert "Page two content" in text
+
+    def test_metadata_extraction(self) -> None:
+        _, pdf_metadata, _ = read_pdf_file(_load("with_metadata.pdf"))
+        assert pdf_metadata.get("Title") == "My Title"
+        assert pdf_metadata.get("Author") == "Jane Doe"
+
+    def test_encrypted_pdf_with_correct_password(self) -> None:
+        text, _, _ = read_pdf_file(_load("encrypted.pdf"), pdf_pass="pass123")
+        assert "Secret Content" in text
+
+    def test_encrypted_pdf_without_password(self) -> None:
+        text, _, _ = read_pdf_file(_load("encrypted.pdf"))
+        assert text == ""
+
+    def test_encrypted_pdf_with_wrong_password(self) -> None:
+        text, _, _ = read_pdf_file(_load("encrypted.pdf"), pdf_pass="wrong")
+        assert text == ""
+
+    def test_empty_pdf(self) -> None:
+        text, _, _ = read_pdf_file(_load("empty.pdf"))
+        assert text.strip() == ""
+
+    def test_invalid_pdf_returns_empty(self) -> None:
+        text, _, images = read_pdf_file(BytesIO(b"this is not a pdf"))
+        assert text == ""
+        assert images == []
+
+    def test_image_extraction_disabled_by_default(self) -> None:
+        _, _, images = read_pdf_file(_load("with_image.pdf"))
+        assert images == []
+
+    def test_image_extraction_collects_images(self) -> None:
+        _, _, images = read_pdf_file(_load("with_image.pdf"), extract_images=True)
+        assert len(images) == 1
+        img_bytes, img_name = images[0]
+        assert len(img_bytes) > 0
+        assert img_name  # non-empty name
+
+    def test_image_callback_streams_instead_of_collecting(self) -> None:
+        """With image_callback, images are streamed via callback and not accumulated."""
+        collected: list[tuple[bytes, str]] = []
+
+        def callback(data: bytes, name: str) -> None:
+            collected.append((data, name))
+
+        _, _, images = read_pdf_file(
+            _load("with_image.pdf"), extract_images=True, image_callback=callback
+        )
+        # Callback received the image
+        assert len(collected) == 1
+        assert len(collected[0][0]) > 0
+        # Returned list is empty when callback is used
+        assert images == []
+
+
+# ── pdf_to_text ──────────────────────────────────────────────────────────
+
+
+class TestPdfToText:
+    def test_returns_text(self) -> None:
+        assert "Hello World" in pdf_to_text(_load("simple.pdf"))
+
+    def test_with_password(self) -> None:
+        assert "Secret Content" in pdf_to_text(
+            _load("encrypted.pdf"), pdf_pass="pass123"
+        )
+
+    def test_encrypted_without_password_returns_empty(self) -> None:
+        assert pdf_to_text(_load("encrypted.pdf")) == ""
+
+
+# ── is_pdf_protected ─────────────────────────────────────────────────────
+
+
+class TestIsPdfProtected:
+    def test_unprotected_pdf(self) -> None:
+        assert is_pdf_protected(_load("simple.pdf")) is False
+
+    def test_protected_pdf(self) -> None:
+        assert is_pdf_protected(_load("encrypted.pdf")) is True
+
+    def test_preserves_file_position(self) -> None:
+        pdf = _load("simple.pdf")
+        pdf.seek(42)
+        is_pdf_protected(pdf)
+        assert pdf.tell() == 42

backend/tests/unit/onyx/hooks/test_base_spec.py

@@ -0,0 +1,19 @@
+import pytest
+from pydantic import BaseModel
+
+from onyx.db.enums import HookPoint
+from onyx.hooks.points.base import HookPointSpec
+
+
+def test_init_subclass_raises_for_missing_attrs() -> None:
+    with pytest.raises(TypeError, match="must define class attributes"):
+
+        class IncompleteSpec(HookPointSpec):
+            hook_point = HookPoint.QUERY_PROCESSING
+            # missing display_name, description, payload_model, response_model, etc.
+
+            class _Payload(BaseModel):
+                pass
+
+            payload_model = _Payload
+            response_model = _Payload

backend/tests/unit/onyx/hooks/test_executor.py

@@ -0,0 +1,678 @@
+"""Unit tests for the hook executor."""
+
+import json
+from typing import Any
+from unittest.mock import MagicMock
+from unittest.mock import patch
+
+import httpx
+import pytest
+from pydantic import BaseModel
+
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
+from onyx.hooks.executor import execute_hook
+from onyx.hooks.executor import HookSkipped
+from onyx.hooks.executor import HookSoftFailed
+from onyx.hooks.points.query_processing import QueryProcessingResponse
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+_PAYLOAD: dict[str, Any] = {"query": "test", "user_email": "u@example.com"}
+# A valid QueryProcessingResponse payload — used by success-path tests.
+_RESPONSE_PAYLOAD: dict[str, Any] = {"query": "better test"}
+
+
+def _make_hook(
+    *,
+    is_active: bool = True,
+    endpoint_url: str | None = "https://hook.example.com/query",
+    api_key: MagicMock | None = None,
+    timeout_seconds: float = 5.0,
+    fail_strategy: HookFailStrategy = HookFailStrategy.SOFT,
+    hook_id: int = 1,
+    is_reachable: bool | None = None,
+    hook_point: HookPoint = HookPoint.QUERY_PROCESSING,
+) -> MagicMock:
+    hook = MagicMock()
+    hook.is_active = is_active
+    hook.endpoint_url = endpoint_url
+    hook.api_key = api_key
+    hook.timeout_seconds = timeout_seconds
+    hook.id = hook_id
+    hook.fail_strategy = fail_strategy
+    hook.is_reachable = is_reachable
+    hook.hook_point = hook_point
+    return hook
+
+
+def _make_api_key(value: str) -> MagicMock:
+    api_key = MagicMock()
+    api_key.get_value.return_value = value
+    return api_key
+
+
+def _make_response(
+    *,
+    status_code: int = 200,
+    json_return: Any = _RESPONSE_PAYLOAD,
+    json_side_effect: Exception | None = None,
+) -> MagicMock:
+    """Build a response mock with controllable json() behaviour."""
+    response = MagicMock()
+    response.status_code = status_code
+    if json_side_effect is not None:
+        response.json.side_effect = json_side_effect
+    else:
+        response.json.return_value = json_return
+    return response
+
+
+def _setup_client(
+    mock_client_cls: MagicMock,
+    *,
+    response: MagicMock | None = None,
+    side_effect: Exception | None = None,
+) -> MagicMock:
+    """Wire up the httpx.Client mock and return the inner client.
+
+    If side_effect is an httpx.HTTPStatusError, it is raised from
+    raise_for_status() (matching real httpx behaviour) and post() returns a
+    response mock with the matching status_code set.  All other exceptions are
+    raised directly from post().
+    """
+    mock_client = MagicMock()
+
+    if isinstance(side_effect, httpx.HTTPStatusError):
+        error_response = MagicMock()
+        error_response.status_code = side_effect.response.status_code
+        error_response.raise_for_status.side_effect = side_effect
+        mock_client.post = MagicMock(return_value=error_response)
+    else:
+        mock_client.post = MagicMock(
+            side_effect=side_effect, return_value=response if not side_effect else None
+        )
+
+    mock_client_cls.return_value.__enter__ = MagicMock(return_value=mock_client)
+    mock_client_cls.return_value.__exit__ = MagicMock(return_value=False)
+    return mock_client
+
+
+# ---------------------------------------------------------------------------
+# Fixtures
+# ---------------------------------------------------------------------------
+
+
+@pytest.fixture()
+def db_session() -> MagicMock:
+    return MagicMock()
+
+
+# ---------------------------------------------------------------------------
+# Early-exit guards (no HTTP call, no DB writes)
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize(
+    "hooks_available,hook",
+    [
+        # HOOKS_AVAILABLE=False exits before the DB lookup — hook is irrelevant.
+        pytest.param(False, None, id="hooks_not_available"),
+        pytest.param(True, None, id="hook_not_found"),
+        pytest.param(True, _make_hook(is_active=False), id="hook_inactive"),
+        pytest.param(True, _make_hook(endpoint_url=None), id="no_endpoint_url"),
+    ],
+)
+def test_early_exit_returns_skipped_with_no_db_writes(
+    db_session: MagicMock,
+    hooks_available: bool,
+    hook: MagicMock | None,
+) -> None:
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", hooks_available),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.update_hook__no_commit") as mock_update,
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit") as mock_log,
+    ):
+        result = execute_hook(
+            db_session=db_session,
+            hook_point=HookPoint.QUERY_PROCESSING,
+            payload=_PAYLOAD,
+        )
+
+    assert isinstance(result, HookSkipped)
+    mock_update.assert_not_called()
+    mock_log.assert_not_called()
+
+
+# ---------------------------------------------------------------------------
+# Successful HTTP call
+# ---------------------------------------------------------------------------
+
+
+def test_success_returns_validated_model_and_sets_reachable(
+    db_session: MagicMock,
+) -> None:
+    hook = _make_hook()
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.get_session_with_current_tenant"),
+        patch("onyx.hooks.executor.update_hook__no_commit") as mock_update,
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit") as mock_log,
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        _setup_client(mock_client_cls, response=_make_response())
+        result = execute_hook(
+            db_session=db_session,
+            hook_point=HookPoint.QUERY_PROCESSING,
+            payload=_PAYLOAD,
+        )
+
+    assert isinstance(result, QueryProcessingResponse)
+    assert result.query == _RESPONSE_PAYLOAD["query"]
+    _, update_kwargs = mock_update.call_args
+    assert update_kwargs["is_reachable"] is True
+    mock_log.assert_not_called()
+
+
+def test_success_skips_reachable_write_when_already_true(db_session: MagicMock) -> None:
+    """Deduplication guard: a hook already at is_reachable=True that succeeds
+    must not trigger a DB write."""
+    hook = _make_hook(is_reachable=True)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.get_session_with_current_tenant"),
+        patch("onyx.hooks.executor.update_hook__no_commit") as mock_update,
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit"),
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        _setup_client(mock_client_cls, response=_make_response())
+        result = execute_hook(
+            db_session=db_session,
+            hook_point=HookPoint.QUERY_PROCESSING,
+            payload=_PAYLOAD,
+        )
+
+    assert isinstance(result, QueryProcessingResponse)
+    assert result.query == _RESPONSE_PAYLOAD["query"]
+    mock_update.assert_not_called()
+
+
+def test_non_dict_json_response_is_a_failure(db_session: MagicMock) -> None:
+    """response.json() returning a non-dict (e.g. list) must be treated as failure.
+    The server responded, so is_reachable is not updated."""
+    hook = _make_hook(fail_strategy=HookFailStrategy.SOFT)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.get_session_with_current_tenant"),
+        patch("onyx.hooks.executor.update_hook__no_commit") as mock_update,
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit") as mock_log,
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        _setup_client(
+            mock_client_cls,
+            response=_make_response(json_return=["unexpected", "list"]),
+        )
+        result = execute_hook(
+            db_session=db_session,
+            hook_point=HookPoint.QUERY_PROCESSING,
+            payload=_PAYLOAD,
+        )
+
+    assert isinstance(result, HookSoftFailed)
+    _, log_kwargs = mock_log.call_args
+    assert log_kwargs["is_success"] is False
+    assert "non-dict" in (log_kwargs["error_message"] or "")
+    mock_update.assert_not_called()
+
+
+def test_json_decode_failure_is_a_failure(db_session: MagicMock) -> None:
+    """response.json() raising must be treated as failure with SOFT strategy.
+    The server responded, so is_reachable is not updated."""
+    hook = _make_hook(fail_strategy=HookFailStrategy.SOFT)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.get_session_with_current_tenant"),
+        patch("onyx.hooks.executor.update_hook__no_commit") as mock_update,
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit") as mock_log,
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        _setup_client(
+            mock_client_cls,
+            response=_make_response(
+                json_side_effect=json.JSONDecodeError("not JSON", "", 0)
+            ),
+        )
+        result = execute_hook(
+            db_session=db_session,
+            hook_point=HookPoint.QUERY_PROCESSING,
+            payload=_PAYLOAD,
+        )
+
+    assert isinstance(result, HookSoftFailed)
+    _, log_kwargs = mock_log.call_args
+    assert log_kwargs["is_success"] is False
+    assert "non-JSON" in (log_kwargs["error_message"] or "")
+    mock_update.assert_not_called()
+
+
+# ---------------------------------------------------------------------------
+# HTTP failure paths
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize(
+    "exception,fail_strategy,expected_type,expected_is_reachable",
+    [
+        # NetworkError → is_reachable=False
+        pytest.param(
+            httpx.ConnectError("refused"),
+            HookFailStrategy.SOFT,
+            HookSoftFailed,
+            False,
+            id="connect_error_soft",
+        ),
+        pytest.param(
+            httpx.ConnectError("refused"),
+            HookFailStrategy.HARD,
+            OnyxError,
+            False,
+            id="connect_error_hard",
+        ),
+        # 401/403 → is_reachable=False (api_key revoked)
+        pytest.param(
+            httpx.HTTPStatusError(
+                "401",
+                request=MagicMock(),
+                response=MagicMock(status_code=401, text="Unauthorized"),
+            ),
+            HookFailStrategy.SOFT,
+            HookSoftFailed,
+            False,
+            id="auth_401_soft",
+        ),
+        pytest.param(
+            httpx.HTTPStatusError(
+                "403",
+                request=MagicMock(),
+                response=MagicMock(status_code=403, text="Forbidden"),
+            ),
+            HookFailStrategy.HARD,
+            OnyxError,
+            False,
+            id="auth_403_hard",
+        ),
+        # TimeoutException → no is_reachable write (None)
+        pytest.param(
+            httpx.TimeoutException("timeout"),
+            HookFailStrategy.SOFT,
+            HookSoftFailed,
+            None,
+            id="timeout_soft",
+        ),
+        pytest.param(
+            httpx.TimeoutException("timeout"),
+            HookFailStrategy.HARD,
+            OnyxError,
+            None,
+            id="timeout_hard",
+        ),
+        # Other HTTP errors → no is_reachable write (None)
+        pytest.param(
+            httpx.HTTPStatusError(
+                "500",
+                request=MagicMock(),
+                response=MagicMock(status_code=500, text="error"),
+            ),
+            HookFailStrategy.SOFT,
+            HookSoftFailed,
+            None,
+            id="http_status_error_soft",
+        ),
+        pytest.param(
+            httpx.HTTPStatusError(
+                "500",
+                request=MagicMock(),
+                response=MagicMock(status_code=500, text="error"),
+            ),
+            HookFailStrategy.HARD,
+            OnyxError,
+            None,
+            id="http_status_error_hard",
+        ),
+    ],
+)
+def test_http_failure_paths(
+    db_session: MagicMock,
+    exception: Exception,
+    fail_strategy: HookFailStrategy,
+    expected_type: type,
+    expected_is_reachable: bool | None,
+) -> None:
+    hook = _make_hook(fail_strategy=fail_strategy)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.get_session_with_current_tenant"),
+        patch("onyx.hooks.executor.update_hook__no_commit") as mock_update,
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit"),
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        _setup_client(mock_client_cls, side_effect=exception)
+
+        if expected_type is OnyxError:
+            with pytest.raises(OnyxError) as exc_info:
+                execute_hook(
+                    db_session=db_session,
+                    hook_point=HookPoint.QUERY_PROCESSING,
+                    payload=_PAYLOAD,
+                )
+            assert exc_info.value.error_code is OnyxErrorCode.HOOK_EXECUTION_FAILED
+        else:
+            result = execute_hook(
+                db_session=db_session,
+                hook_point=HookPoint.QUERY_PROCESSING,
+                payload=_PAYLOAD,
+            )
+            assert isinstance(result, expected_type)
+
+    if expected_is_reachable is None:
+        mock_update.assert_not_called()
+    else:
+        mock_update.assert_called_once()
+        _, kwargs = mock_update.call_args
+        assert kwargs["is_reachable"] is expected_is_reachable
+
+
+# ---------------------------------------------------------------------------
+# Authorization header
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize(
+    "api_key_value,expect_auth_header",
+    [
+        pytest.param("secret-token", True, id="api_key_present"),
+        pytest.param(None, False, id="api_key_absent"),
+    ],
+)
+def test_authorization_header(
+    db_session: MagicMock,
+    api_key_value: str | None,
+    expect_auth_header: bool,
+) -> None:
+    api_key = _make_api_key(api_key_value) if api_key_value else None
+    hook = _make_hook(api_key=api_key)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.get_session_with_current_tenant"),
+        patch("onyx.hooks.executor.update_hook__no_commit"),
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit"),
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        mock_client = _setup_client(mock_client_cls, response=_make_response())
+        execute_hook(
+            db_session=db_session,
+            hook_point=HookPoint.QUERY_PROCESSING,
+            payload=_PAYLOAD,
+        )
+
+    _, call_kwargs = mock_client.post.call_args
+    if expect_auth_header:
+        assert call_kwargs["headers"]["Authorization"] == f"Bearer {api_key_value}"
+    else:
+        assert "Authorization" not in call_kwargs["headers"]
+
+
+# ---------------------------------------------------------------------------
+# Persist session failure
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize(
+    "http_exception,expect_onyx_error",
+    [
+        pytest.param(None, False, id="success_path"),
+        pytest.param(httpx.ConnectError("refused"), True, id="hard_fail_path"),
+    ],
+)
+def test_persist_session_failure_is_swallowed(
+    db_session: MagicMock,
+    http_exception: Exception | None,
+    expect_onyx_error: bool,
+) -> None:
+    """DB session failure in _persist_result must not mask the real return value or OnyxError."""
+    hook = _make_hook(fail_strategy=HookFailStrategy.HARD)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch(
+            "onyx.hooks.executor.get_session_with_current_tenant",
+            side_effect=RuntimeError("DB unavailable"),
+        ),
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        _setup_client(
+            mock_client_cls,
+            response=_make_response() if not http_exception else None,
+            side_effect=http_exception,
+        )
+
+        if expect_onyx_error:
+            with pytest.raises(OnyxError) as exc_info:
+                execute_hook(
+                    db_session=db_session,
+                    hook_point=HookPoint.QUERY_PROCESSING,
+                    payload=_PAYLOAD,
+                )
+            assert exc_info.value.error_code is OnyxErrorCode.HOOK_EXECUTION_FAILED
+        else:
+            result = execute_hook(
+                db_session=db_session,
+                hook_point=HookPoint.QUERY_PROCESSING,
+                payload=_PAYLOAD,
+            )
+            assert isinstance(result, QueryProcessingResponse)
+            assert result.query == _RESPONSE_PAYLOAD["query"]
+
+
+# ---------------------------------------------------------------------------
+# Response model validation
+# ---------------------------------------------------------------------------
+
+
+class _StrictResponse(BaseModel):
+    """Strict model used to reliably trigger a ValidationError in tests."""
+
+    required_field: str  # no default → missing key raises ValidationError
+
+
+def _make_strict_spec() -> MagicMock:
+    spec = MagicMock()
+    spec.response_model = _StrictResponse
+    return spec
+
+
+@pytest.mark.parametrize(
+    "fail_strategy,expected_type",
+    [
+        pytest.param(
+            HookFailStrategy.SOFT, HookSoftFailed, id="validation_failure_soft"
+        ),
+        pytest.param(HookFailStrategy.HARD, OnyxError, id="validation_failure_hard"),
+    ],
+)
+def test_response_validation_failure_respects_fail_strategy(
+    db_session: MagicMock,
+    fail_strategy: HookFailStrategy,
+    expected_type: type,
+) -> None:
+    """A response that fails response_model validation is treated like any other
+    hook failure: logged, is_reachable left unchanged, fail_strategy respected."""
+    hook = _make_hook(fail_strategy=fail_strategy)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.get_session_with_current_tenant"),
+        patch("onyx.hooks.executor.update_hook__no_commit") as mock_update,
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit") as mock_log,
+        patch(
+            "onyx.hooks.executor.get_hook_point_spec",
+            return_value=_make_strict_spec(),
+        ),
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        # Response payload is missing required_field → ValidationError
+        _setup_client(mock_client_cls, response=_make_response(json_return={}))
+
+        if expected_type is OnyxError:
+            with pytest.raises(OnyxError) as exc_info:
+                execute_hook(
+                    db_session=db_session,
+                    hook_point=HookPoint.QUERY_PROCESSING,
+                    payload=_PAYLOAD,
+                )
+            assert exc_info.value.error_code is OnyxErrorCode.HOOK_EXECUTION_FAILED
+        else:
+            result = execute_hook(
+                db_session=db_session,
+                hook_point=HookPoint.QUERY_PROCESSING,
+                payload=_PAYLOAD,
+            )
+            assert isinstance(result, HookSoftFailed)
+
+    # is_reachable must not be updated — server responded correctly
+    mock_update.assert_not_called()
+    # failure must be logged
+    mock_log.assert_called_once()
+    _, log_kwargs = mock_log.call_args
+    assert log_kwargs["is_success"] is False
+    assert "validation" in (log_kwargs["error_message"] or "").lower()
+
+
+# ---------------------------------------------------------------------------
+# Outer soft-fail guard in execute_hook
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.parametrize(
+    "fail_strategy,expected_type",
+    [
+        pytest.param(HookFailStrategy.SOFT, HookSoftFailed, id="unexpected_exc_soft"),
+        pytest.param(HookFailStrategy.HARD, ValueError, id="unexpected_exc_hard"),
+    ],
+)
+def test_unexpected_exception_in_inner_respects_fail_strategy(
+    db_session: MagicMock,
+    fail_strategy: HookFailStrategy,
+    expected_type: type,
+) -> None:
+    """An unexpected exception raised by _execute_hook_inner (not an OnyxError from
+    HARD fail — e.g. a bug or an assertion error) must be swallowed and return
+    HookSoftFailed for SOFT strategy, or re-raised for HARD strategy."""
+    hook = _make_hook(fail_strategy=fail_strategy)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch(
+            "onyx.hooks.executor._execute_hook_inner",
+            side_effect=ValueError("unexpected bug"),
+        ),
+    ):
+        if expected_type is HookSoftFailed:
+            result = execute_hook(
+                db_session=db_session,
+                hook_point=HookPoint.QUERY_PROCESSING,
+                payload=_PAYLOAD,
+            )
+            assert isinstance(result, HookSoftFailed)
+        else:
+            with pytest.raises(ValueError, match="unexpected bug"):
+                execute_hook(
+                    db_session=db_session,
+                    hook_point=HookPoint.QUERY_PROCESSING,
+                    payload=_PAYLOAD,
+                )
+
+
+def test_is_reachable_failure_does_not_prevent_log(db_session: MagicMock) -> None:
+    """is_reachable update failing (e.g. concurrent hook deletion) must not
+    prevent the execution log from being written.
+
+    Simulates the production failure path: update_hook__no_commit raises
+    OnyxError(NOT_FOUND) as it would if the hook was concurrently deleted
+    between the initial lookup and the reachable update.
+    """
+    hook = _make_hook(fail_strategy=HookFailStrategy.SOFT)
+
+    with (
+        patch("onyx.hooks.executor.HOOKS_AVAILABLE", True),
+        patch(
+            "onyx.hooks.executor.get_non_deleted_hook_by_hook_point",
+            return_value=hook,
+        ),
+        patch("onyx.hooks.executor.get_session_with_current_tenant"),
+        patch(
+            "onyx.hooks.executor.update_hook__no_commit",
+            side_effect=OnyxError(OnyxErrorCode.NOT_FOUND, "hook deleted"),
+        ),
+        patch("onyx.hooks.executor.create_hook_execution_log__no_commit") as mock_log,
+        patch("httpx.Client") as mock_client_cls,
+    ):
+        _setup_client(mock_client_cls, side_effect=httpx.ConnectError("refused"))
+        result = execute_hook(
+            db_session=db_session,
+            hook_point=HookPoint.QUERY_PROCESSING,
+            payload=_PAYLOAD,
+        )
+
+    assert isinstance(result, HookSoftFailed)
+    mock_log.assert_called_once()

backend/tests/unit/onyx/hooks/test_models.py

@@ -0,0 +1,86 @@
+import pytest
+from pydantic import ValidationError
+
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.hooks.models import HookCreateRequest
+from onyx.hooks.models import HookUpdateRequest
+
+
+def test_hook_update_request_rejects_empty() -> None:
+    # No fields supplied at all
+    with pytest.raises(ValidationError, match="At least one field must be provided"):
+        HookUpdateRequest()
+
+
+def test_hook_update_request_rejects_null_name_when_only_field() -> None:
+    # Explicitly setting name=None is rejected as name cannot be cleared
+    with pytest.raises(ValidationError, match="name cannot be cleared"):
+        HookUpdateRequest(name=None)
+
+
+def test_hook_update_request_accepts_single_field() -> None:
+    req = HookUpdateRequest(name="new name")
+    assert req.name == "new name"
+
+
+def test_hook_update_request_accepts_partial_fields() -> None:
+    req = HookUpdateRequest(fail_strategy=HookFailStrategy.SOFT, timeout_seconds=10.0)
+    assert req.fail_strategy == HookFailStrategy.SOFT
+    assert req.timeout_seconds == 10.0
+    assert req.name is None
+
+
+def test_hook_update_request_rejects_null_name() -> None:
+    with pytest.raises(ValidationError, match="name cannot be cleared"):
+        HookUpdateRequest(name=None, fail_strategy=HookFailStrategy.SOFT)
+
+
+def test_hook_update_request_rejects_empty_name() -> None:
+    with pytest.raises(ValidationError, match="name cannot be cleared"):
+        HookUpdateRequest(name="", fail_strategy=HookFailStrategy.SOFT)
+
+
+def test_hook_update_request_rejects_null_endpoint_url() -> None:
+    with pytest.raises(ValidationError, match="endpoint_url cannot be cleared"):
+        HookUpdateRequest(endpoint_url=None, fail_strategy=HookFailStrategy.SOFT)
+
+
+def test_hook_update_request_rejects_empty_endpoint_url() -> None:
+    with pytest.raises(ValidationError, match="endpoint_url cannot be cleared"):
+        HookUpdateRequest(endpoint_url="", fail_strategy=HookFailStrategy.SOFT)
+
+
+def test_hook_update_request_allows_null_api_key() -> None:
+    # api_key=null is valid — means "clear the api key"
+    req = HookUpdateRequest(api_key=None)
+    assert req.api_key is None
+    assert "api_key" in req.model_fields_set
+
+
+def test_hook_update_request_rejects_whitespace_name() -> None:
+    with pytest.raises(ValidationError, match="name cannot be cleared"):
+        HookUpdateRequest(name="   ", fail_strategy=HookFailStrategy.SOFT)
+
+
+def test_hook_update_request_rejects_whitespace_endpoint_url() -> None:
+    with pytest.raises(ValidationError, match="endpoint_url cannot be cleared"):
+        HookUpdateRequest(endpoint_url="   ", fail_strategy=HookFailStrategy.SOFT)
+
+
+def test_hook_create_request_rejects_whitespace_name() -> None:
+    with pytest.raises(ValidationError, match="whitespace-only"):
+        HookCreateRequest(
+            name="   ",
+            hook_point=HookPoint.QUERY_PROCESSING,
+            endpoint_url="https://example.com/hook",
+        )
+
+
+def test_hook_create_request_rejects_whitespace_endpoint_url() -> None:
+    with pytest.raises(ValidationError, match="whitespace-only"):
+        HookCreateRequest(
+            name="my hook",
+            hook_point=HookPoint.QUERY_PROCESSING,
+            endpoint_url="   ",
+        )

backend/tests/unit/onyx/hooks/test_query_processing_spec.py

@@ -0,0 +1,62 @@
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.hooks.points.query_processing import QueryProcessingSpec
+
+
+def test_hook_point_is_query_processing() -> None:
+    assert QueryProcessingSpec().hook_point == HookPoint.QUERY_PROCESSING
+
+
+def test_default_fail_strategy_is_hard() -> None:
+    assert QueryProcessingSpec().default_fail_strategy == HookFailStrategy.HARD
+
+
+def test_default_timeout_seconds() -> None:
+    # User is actively waiting — 5s is the documented contract for this hook point
+    assert QueryProcessingSpec().default_timeout_seconds == 5.0
+
+
+def test_input_schema_required_fields() -> None:
+    schema = QueryProcessingSpec().input_schema
+    assert schema["type"] == "object"
+    required = schema["required"]
+    assert "query" in required
+    assert "user_email" in required
+    assert "chat_session_id" in required
+
+
+def test_input_schema_chat_session_id_is_string() -> None:
+    props = QueryProcessingSpec().input_schema["properties"]
+    assert props["chat_session_id"]["type"] == "string"
+
+
+def test_input_schema_query_is_string() -> None:
+    props = QueryProcessingSpec().input_schema["properties"]
+    assert props["query"]["type"] == "string"
+
+
+def test_input_schema_user_email_is_nullable() -> None:
+    props = QueryProcessingSpec().input_schema["properties"]
+    # Pydantic v2 emits anyOf for nullable fields
+    assert any(s.get("type") == "null" for s in props["user_email"]["anyOf"])
+
+
+def test_output_schema_query_is_optional() -> None:
+    # query defaults to None (absent = reject); not required in the schema
+    schema = QueryProcessingSpec().output_schema
+    assert "query" not in schema.get("required", [])
+
+
+def test_output_schema_query_is_nullable() -> None:
+    # null means "reject the query"; Pydantic v2 emits anyOf for nullable fields
+    props = QueryProcessingSpec().output_schema["properties"]
+    assert any(s.get("type") == "null" for s in props["query"]["anyOf"])
+
+
+def test_output_schema_rejection_message_is_optional() -> None:
+    schema = QueryProcessingSpec().output_schema
+    assert "rejection_message" not in schema.get("required", [])
+
+
+def test_input_schema_no_additional_properties() -> None:
+    assert QueryProcessingSpec().input_schema.get("additionalProperties") is False

backend/tests/unit/onyx/hooks/test_registry.py

@@ -0,0 +1,47 @@
+import pytest
+
+from onyx.db.enums import HookPoint
+from onyx.hooks import registry as registry_module
+from onyx.hooks.registry import get_all_specs
+from onyx.hooks.registry import get_hook_point_spec
+from onyx.hooks.registry import validate_registry
+
+
+def test_registry_covers_all_hook_points() -> None:
+    """Every HookPoint enum member must have a registered spec."""
+    assert {s.hook_point for s in get_all_specs()} == set(
+        HookPoint
+    ), f"Missing specs for: {set(HookPoint) - {s.hook_point for s in get_all_specs()}}"
+
+
+def test_get_hook_point_spec_returns_correct_spec() -> None:
+    for hook_point in HookPoint:
+        spec = get_hook_point_spec(hook_point)
+        assert spec.hook_point == hook_point
+
+
+def test_get_all_specs_returns_all() -> None:
+    specs = get_all_specs()
+    assert len(specs) == len(HookPoint)
+    assert {s.hook_point for s in specs} == set(HookPoint)
+
+
+def test_get_hook_point_spec_raises_for_unregistered(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """get_hook_point_spec raises ValueError when a hook point has no spec."""
+    monkeypatch.setattr(registry_module, "_REGISTRY", {})
+    with pytest.raises(ValueError, match="No spec registered for hook point"):
+        get_hook_point_spec(HookPoint.QUERY_PROCESSING)
+
+
+def test_validate_registry_passes() -> None:
+    validate_registry()  # should not raise with the real registry
+
+
+def test_validate_registry_raises_for_incomplete(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(registry_module, "_REGISTRY", {})
+    with pytest.raises(RuntimeError, match="Hook point\\(s\\) have no registered spec"):
+        validate_registry()

backend/tests/unit/onyx/llm/test_multi_llm.py

@@ -256,7 +256,6 @@ def test_multiple_tool_calls(default_multi_llm: LitellmLLM) -> None:
                 {"role": "user", "content": "What's the weather and time in New York?"}
             ],
             tools=tools,
-            tool_choice=None,
             stream=True,
             temperature=0.0,  # Default value from GEN_AI_TEMPERATURE
             timeout=30,
@@ -412,7 +411,6 @@ def test_multiple_tool_calls_streaming(default_multi_llm: LitellmLLM) -> None:
                 {"role": "user", "content": "What's the weather and time in New York?"}
             ],
             tools=tools,
-            tool_choice=None,
             stream=True,
             temperature=0.0,  # Default value from GEN_AI_TEMPERATURE
             timeout=30,
@@ -1431,3 +1429,36 @@ def test_strip_tool_content_merges_consecutive_tool_results() -> None:
     assert "sunny 72F" in merged
     assert "tc_2" in merged
     assert "headline news" in merged
+
+
+def test_no_tool_choice_sent_when_no_tools(default_multi_llm: LitellmLLM) -> None:
+    """Regression test for providers (e.g. Fireworks) that reject tool_choice=null.
+
+    When no tools are provided, tool_choice must not be forwarded to
+    litellm.completion() at all — not even as None.
+    """
+    messages: LanguageModelInput = [UserMessage(content="Hello!")]
+
+    mock_stream_chunks = [
+        litellm.ModelResponse(
+            id="chatcmpl-123",
+            choices=[
+                litellm.Choices(
+                    delta=_create_delta(role="assistant", content="Hello!"),
+                    finish_reason="stop",
+                    index=0,
+                )
+            ],
+            model="gpt-3.5-turbo",
+        ),
+    ]
+
+    with patch("litellm.completion") as mock_completion:
+        mock_completion.return_value = mock_stream_chunks
+
+        default_multi_llm.invoke(messages, tools=None)
+
+        _, kwargs = mock_completion.call_args
+        assert (
+            "tool_choice" not in kwargs
+        ), "tool_choice must not be sent to providers when no tools are provided"

backend/tests/unit/onyx/llm/test_vision_model_selection_logging.py

@@ -0,0 +1,130 @@
+"""
+Unit tests for vision model selection logging in get_default_llm_with_vision.
+
+Verifies that operators get clear feedback about:
+1. Which vision model was selected and why
+2. When the default vision model doesn't support image input
+3. When no vision-capable model exists at all
+"""
+
+from unittest.mock import MagicMock
+from unittest.mock import patch
+
+from onyx.llm.factory import get_default_llm_with_vision
+
+
+_FACTORY = "onyx.llm.factory"
+
+
+def _make_mock_model(
+    *,
+    name: str = "gpt-4o",
+    provider: str = "openai",
+    provider_id: int = 1,
+    flow_types: list[str] | None = None,
+) -> MagicMock:
+    model = MagicMock()
+    model.name = name
+    model.llm_provider_id = provider_id
+    model.llm_provider.provider = provider
+    model.llm_model_flow_types = flow_types or []
+    return model
+
+
+@patch(f"{_FACTORY}.get_session_with_current_tenant")
+@patch(f"{_FACTORY}.fetch_default_vision_model")
+@patch(f"{_FACTORY}.model_supports_image_input", return_value=True)
+@patch(f"{_FACTORY}.llm_from_provider")
+@patch(f"{_FACTORY}.LLMProviderView")
+@patch(f"{_FACTORY}.logger")
+def test_logs_when_using_default_vision_model(
+    mock_logger: MagicMock,
+    mock_provider_view: MagicMock,  # noqa: ARG001
+    mock_llm_from: MagicMock,  # noqa: ARG001
+    mock_supports: MagicMock,  # noqa: ARG001
+    mock_fetch_default: MagicMock,
+    mock_session: MagicMock,  # noqa: ARG001
+) -> None:
+    mock_fetch_default.return_value = _make_mock_model(name="gpt-4o", provider="azure")
+
+    get_default_llm_with_vision()
+
+    mock_logger.info.assert_called_once()
+    log_msg = mock_logger.info.call_args[0][0]
+    assert "default vision model" in log_msg.lower()
+
+
+@patch(f"{_FACTORY}.get_session_with_current_tenant")
+@patch(f"{_FACTORY}.fetch_default_vision_model")
+@patch(f"{_FACTORY}.model_supports_image_input", return_value=False)
+@patch(f"{_FACTORY}.fetch_existing_models", return_value=[])
+@patch(f"{_FACTORY}.logger")
+def test_warns_when_default_model_lacks_vision(
+    mock_logger: MagicMock,
+    mock_fetch_models: MagicMock,  # noqa: ARG001
+    mock_supports: MagicMock,  # noqa: ARG001
+    mock_fetch_default: MagicMock,
+    mock_session: MagicMock,  # noqa: ARG001
+) -> None:
+    mock_fetch_default.return_value = _make_mock_model(
+        name="text-only-model", provider="azure"
+    )
+
+    result = get_default_llm_with_vision()
+
+    assert result is None
+    # Should have warned about the default model not supporting vision
+    warning_calls = [
+        call
+        for call in mock_logger.warning.call_args_list
+        if "does not support" in str(call)
+    ]
+    assert len(warning_calls) >= 1
+
+
+@patch(f"{_FACTORY}.get_session_with_current_tenant")
+@patch(f"{_FACTORY}.fetch_default_vision_model", return_value=None)
+@patch(f"{_FACTORY}.fetch_existing_models", return_value=[])
+@patch(f"{_FACTORY}.logger")
+def test_warns_when_no_models_exist(
+    mock_logger: MagicMock,
+    mock_fetch_models: MagicMock,  # noqa: ARG001
+    mock_fetch_default: MagicMock,  # noqa: ARG001
+    mock_session: MagicMock,  # noqa: ARG001
+) -> None:
+    result = get_default_llm_with_vision()
+
+    assert result is None
+    mock_logger.warning.assert_called_once()
+    log_msg = mock_logger.warning.call_args[0][0]
+    assert "no llm models" in log_msg.lower()
+
+
+@patch(f"{_FACTORY}.get_session_with_current_tenant")
+@patch(f"{_FACTORY}.fetch_default_vision_model", return_value=None)
+@patch(f"{_FACTORY}.fetch_existing_models")
+@patch(f"{_FACTORY}.model_supports_image_input", return_value=False)
+@patch(f"{_FACTORY}.LLMProviderView")
+@patch(f"{_FACTORY}.logger")
+def test_warns_when_no_model_supports_vision(
+    mock_logger: MagicMock,
+    mock_provider_view: MagicMock,  # noqa: ARG001
+    mock_supports: MagicMock,  # noqa: ARG001
+    mock_fetch_models: MagicMock,
+    mock_fetch_default: MagicMock,  # noqa: ARG001
+    mock_session: MagicMock,  # noqa: ARG001
+) -> None:
+    mock_fetch_models.return_value = [
+        _make_mock_model(name="text-model-1", provider="openai"),
+        _make_mock_model(name="text-model-2", provider="azure", provider_id=2),
+    ]
+
+    result = get_default_llm_with_vision()
+
+    assert result is None
+    warning_calls = [
+        call
+        for call in mock_logger.warning.call_args_list
+        if "no vision-capable model" in str(call).lower()
+    ]
+    assert len(warning_calls) == 1

backend/tests/unit/onyx/server/features/hooks/test_api.py

@@ -0,0 +1,278 @@
+"""Unit tests for onyx.server.features.hooks.api helpers.
+
+Covers:
+- _check_ssrf_safety: scheme enforcement and private-IP blocklist
+- _validate_endpoint: httpx exception → HookValidateStatus mapping
+  ConnectTimeout     → cannot_connect  (TCP handshake never completed)
+  ConnectError       → cannot_connect  (DNS / TLS failure)
+  ReadTimeout et al. → timeout         (TCP connected, server slow)
+  Any other exc      → cannot_connect
+- _raise_for_validation_failure: HookValidateStatus → OnyxError mapping
+"""
+
+from unittest.mock import MagicMock
+from unittest.mock import patch
+
+import httpx
+import pytest
+
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
+from onyx.hooks.models import HookValidateResponse
+from onyx.hooks.models import HookValidateStatus
+from onyx.server.features.hooks.api import _check_ssrf_safety
+from onyx.server.features.hooks.api import _raise_for_validation_failure
+from onyx.server.features.hooks.api import _validate_endpoint
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+_URL = "https://example.com/hook"
+_API_KEY = "secret"
+_TIMEOUT = 5.0
+
+
+def _mock_response(status_code: int) -> MagicMock:
+    response = MagicMock()
+    response.status_code = status_code
+    return response
+
+
+# ---------------------------------------------------------------------------
+# _check_ssrf_safety
+# ---------------------------------------------------------------------------
+
+
+class TestCheckSsrfSafety:
+    def _call(self, url: str) -> None:
+        _check_ssrf_safety(url)
+
+    # --- scheme checks ---
+
+    def test_https_is_allowed(self) -> None:
+        with patch("onyx.utils.url.socket.getaddrinfo") as mock_dns:
+            mock_dns.return_value = [(None, None, None, None, ("93.184.216.34", 0))]
+            self._call("https://example.com/hook")  # must not raise
+
+    @pytest.mark.parametrize(
+        "url", ["http://example.com/hook", "ftp://example.com/hook"]
+    )
+    def test_non_https_scheme_rejected(self, url: str) -> None:
+        with pytest.raises(OnyxError) as exc_info:
+            self._call(url)
+        assert exc_info.value.error_code == OnyxErrorCode.INVALID_INPUT
+        assert "https" in (exc_info.value.detail or "").lower()
+
+    # --- private IP blocklist ---
+
+    @pytest.mark.parametrize(
+        "ip",
+        [
+            pytest.param("127.0.0.1", id="loopback"),
+            pytest.param("10.0.0.1", id="RFC1918-A"),
+            pytest.param("172.16.0.1", id="RFC1918-B"),
+            pytest.param("192.168.1.1", id="RFC1918-C"),
+            pytest.param("169.254.169.254", id="link-local-IMDS"),
+            pytest.param("100.64.0.1", id="shared-address-space"),
+            pytest.param("::1", id="IPv6-loopback"),
+            pytest.param("fc00::1", id="IPv6-ULA"),
+            pytest.param("fe80::1", id="IPv6-link-local"),
+        ],
+    )
+    def test_private_ip_is_blocked(self, ip: str) -> None:
+        with (
+            patch("onyx.utils.url.socket.getaddrinfo") as mock_dns,
+            pytest.raises(OnyxError) as exc_info,
+        ):
+            mock_dns.return_value = [(None, None, None, None, (ip, 0))]
+            self._call("https://internal.example.com/hook")
+        assert exc_info.value.error_code == OnyxErrorCode.INVALID_INPUT
+        assert ip in (exc_info.value.detail or "")
+
+    def test_public_ip_is_allowed(self) -> None:
+        with patch("onyx.utils.url.socket.getaddrinfo") as mock_dns:
+            mock_dns.return_value = [(None, None, None, None, ("93.184.216.34", 0))]
+            self._call("https://example.com/hook")  # must not raise
+
+    def test_dns_resolution_failure_raises(self) -> None:
+        import socket
+
+        with (
+            patch(
+                "onyx.utils.url.socket.getaddrinfo",
+                side_effect=socket.gaierror("name not found"),
+            ),
+            pytest.raises(OnyxError) as exc_info,
+        ):
+            self._call("https://no-such-host.example.com/hook")
+        assert exc_info.value.error_code == OnyxErrorCode.INVALID_INPUT
+
+
+# ---------------------------------------------------------------------------
+# _validate_endpoint
+# ---------------------------------------------------------------------------
+
+
+class TestValidateEndpoint:
+    def _call(self, *, api_key: str | None = _API_KEY) -> HookValidateResponse:
+        # Bypass SSRF check — tested separately in TestCheckSsrfSafety.
+        with patch("onyx.server.features.hooks.api._check_ssrf_safety"):
+            return _validate_endpoint(
+                endpoint_url=_URL,
+                api_key=api_key,
+                timeout_seconds=_TIMEOUT,
+            )
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    def test_2xx_returns_passed(self, mock_client_cls: MagicMock) -> None:
+        mock_client_cls.return_value.__enter__.return_value.post.return_value = (
+            _mock_response(200)
+        )
+        assert self._call().status == HookValidateStatus.passed
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    def test_5xx_returns_passed(self, mock_client_cls: MagicMock) -> None:
+        mock_client_cls.return_value.__enter__.return_value.post.return_value = (
+            _mock_response(500)
+        )
+        assert self._call().status == HookValidateStatus.passed
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    @pytest.mark.parametrize("status_code", [401, 403])
+    def test_401_403_returns_auth_failed(
+        self, mock_client_cls: MagicMock, status_code: int
+    ) -> None:
+        mock_client_cls.return_value.__enter__.return_value.post.return_value = (
+            _mock_response(status_code)
+        )
+        result = self._call()
+        assert result.status == HookValidateStatus.auth_failed
+        assert str(status_code) in (result.error_message or "")
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    def test_4xx_non_auth_returns_passed(self, mock_client_cls: MagicMock) -> None:
+        mock_client_cls.return_value.__enter__.return_value.post.return_value = (
+            _mock_response(422)
+        )
+        assert self._call().status == HookValidateStatus.passed
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    def test_connect_timeout_returns_cannot_connect(
+        self, mock_client_cls: MagicMock
+    ) -> None:
+        mock_client_cls.return_value.__enter__.return_value.post.side_effect = (
+            httpx.ConnectTimeout("timed out")
+        )
+        assert self._call().status == HookValidateStatus.cannot_connect
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    @pytest.mark.parametrize(
+        "exc",
+        [
+            httpx.ReadTimeout("read timeout"),
+            httpx.WriteTimeout("write timeout"),
+            httpx.PoolTimeout("pool timeout"),
+        ],
+    )
+    def test_read_write_pool_timeout_returns_timeout(
+        self, mock_client_cls: MagicMock, exc: httpx.TimeoutException
+    ) -> None:
+        mock_client_cls.return_value.__enter__.return_value.post.side_effect = exc
+        assert self._call().status == HookValidateStatus.timeout
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    def test_connect_error_returns_cannot_connect(
+        self, mock_client_cls: MagicMock
+    ) -> None:
+        # Covers DNS failures, TLS errors, and other connection-level errors.
+        mock_client_cls.return_value.__enter__.return_value.post.side_effect = (
+            httpx.ConnectError("name resolution failed")
+        )
+        assert self._call().status == HookValidateStatus.cannot_connect
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    def test_arbitrary_exception_returns_cannot_connect(
+        self, mock_client_cls: MagicMock
+    ) -> None:
+        mock_client_cls.return_value.__enter__.return_value.post.side_effect = (
+            ConnectionRefusedError("refused")
+        )
+        assert self._call().status == HookValidateStatus.cannot_connect
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    def test_api_key_sent_as_bearer(self, mock_client_cls: MagicMock) -> None:
+        mock_post = mock_client_cls.return_value.__enter__.return_value.post
+        mock_post.return_value = _mock_response(200)
+        self._call(api_key="mykey")
+        _, kwargs = mock_post.call_args
+        assert kwargs["headers"]["Authorization"] == "Bearer mykey"
+
+    @patch("onyx.server.features.hooks.api.httpx.Client")
+    def test_no_api_key_omits_auth_header(self, mock_client_cls: MagicMock) -> None:
+        mock_post = mock_client_cls.return_value.__enter__.return_value.post
+        mock_post.return_value = _mock_response(200)
+        self._call(api_key=None)
+        _, kwargs = mock_post.call_args
+        assert "Authorization" not in kwargs["headers"]
+
+
+# ---------------------------------------------------------------------------
+# _raise_for_validation_failure
+# ---------------------------------------------------------------------------
+
+
+class TestRaiseForValidationFailure:
+    @pytest.mark.parametrize(
+        "status, expected_code",
+        [
+            (HookValidateStatus.auth_failed, OnyxErrorCode.CREDENTIAL_INVALID),
+            (HookValidateStatus.timeout, OnyxErrorCode.GATEWAY_TIMEOUT),
+            (HookValidateStatus.cannot_connect, OnyxErrorCode.BAD_GATEWAY),
+        ],
+    )
+    def test_raises_correct_error_code(
+        self, status: HookValidateStatus, expected_code: OnyxErrorCode
+    ) -> None:
+        validation = HookValidateResponse(status=status, error_message="some error")
+        with pytest.raises(OnyxError) as exc_info:
+            _raise_for_validation_failure(validation)
+        assert exc_info.value.error_code == expected_code
+
+    def test_auth_failed_passes_error_message_directly(self) -> None:
+        validation = HookValidateResponse(
+            status=HookValidateStatus.auth_failed, error_message="bad credentials"
+        )
+        with pytest.raises(OnyxError) as exc_info:
+            _raise_for_validation_failure(validation)
+        assert exc_info.value.detail == "bad credentials"
+
+    @pytest.mark.parametrize(
+        "status", [HookValidateStatus.timeout, HookValidateStatus.cannot_connect]
+    )
+    def test_timeout_and_cannot_connect_wrap_error_message(
+        self, status: HookValidateStatus
+    ) -> None:
+        validation = HookValidateResponse(status=status, error_message="raw error")
+        with pytest.raises(OnyxError) as exc_info:
+            _raise_for_validation_failure(validation)
+        assert exc_info.value.detail == "Endpoint validation failed: raw error"
+
+
+# ---------------------------------------------------------------------------
+# HookValidateStatus enum string values (API contract)
+# ---------------------------------------------------------------------------
+
+
+class TestHookValidateStatusValues:
+    @pytest.mark.parametrize(
+        "status, expected",
+        [
+            (HookValidateStatus.passed, "passed"),
+            (HookValidateStatus.auth_failed, "auth_failed"),
+            (HookValidateStatus.timeout, "timeout"),
+            (HookValidateStatus.cannot_connect, "cannot_connect"),
+        ],
+    )
+    def test_string_values(self, status: HookValidateStatus, expected: str) -> None:
+        assert status == expected

backend/tests/unit/onyx/server/test_upload_files.py

@@ -0,0 +1,109 @@
+import io
+import zipfile
+from unittest.mock import MagicMock
+from unittest.mock import patch
+from zipfile import BadZipFile
+
+import pytest
+from fastapi import UploadFile
+from starlette.datastructures import Headers
+
+from onyx.configs.constants import FileOrigin
+from onyx.server.documents.connector import upload_files
+
+
+def _create_test_zip() -> bytes:
+    """Create a simple in-memory zip file containing two text files."""
+    buf = io.BytesIO()
+    with zipfile.ZipFile(buf, "w") as zf:
+        zf.writestr("file1.txt", "hello")
+        zf.writestr("file2.txt", "world")
+    return buf.getvalue()
+
+
+def _make_upload_file(content: bytes, filename: str, content_type: str) -> UploadFile:
+    return UploadFile(
+        file=io.BytesIO(content),
+        filename=filename,
+        headers=Headers({"content-type": content_type}),
+    )
+
+
+@patch("onyx.server.documents.connector.get_default_file_store")
+def test_upload_zip_with_unzip_true_extracts_files(
+    mock_get_store: MagicMock,
+) -> None:
+    """When unzip=True (default), a zip upload is extracted into individual files."""
+    mock_store = MagicMock()
+    mock_store.save_file.side_effect = lambda **kwargs: f"id-{kwargs['display_name']}"
+    mock_get_store.return_value = mock_store
+
+    zip_bytes = _create_test_zip()
+    upload = _make_upload_file(zip_bytes, "test.zip", "application/zip")
+
+    result = upload_files([upload], FileOrigin.CONNECTOR)
+
+    # Should have extracted the two individual files, not stored the zip itself
+    assert len(result.file_paths) == 2
+    assert "id-file1.txt" in result.file_paths
+    assert "id-file2.txt" in result.file_paths
+    assert "file1.txt" in result.file_names
+    assert "file2.txt" in result.file_names
+
+
+@patch("onyx.server.documents.connector.get_default_file_store")
+def test_upload_zip_with_unzip_false_stores_zip_as_is(
+    mock_get_store: MagicMock,
+) -> None:
+    """When unzip=False, the zip file is stored as-is without extraction."""
+    mock_store = MagicMock()
+    mock_store.save_file.return_value = "zip-file-id"
+    mock_get_store.return_value = mock_store
+
+    zip_bytes = _create_test_zip()
+    upload = _make_upload_file(zip_bytes, "site_export.zip", "application/zip")
+
+    result = upload_files([upload], FileOrigin.CONNECTOR, unzip=False)
+
+    # Should store exactly one file (the zip itself)
+    assert len(result.file_paths) == 1
+    assert result.file_paths[0] == "zip-file-id"
+    assert result.file_names == ["site_export.zip"]
+    # No zip metadata should be created
+    assert result.zip_metadata_file_id is None
+
+    # Verify the stored content is a valid zip
+    saved_content: io.BytesIO = mock_store.save_file.call_args[1]["content"]
+    saved_content.seek(0)
+    with zipfile.ZipFile(saved_content, "r") as zf:
+        assert set(zf.namelist()) == {"file1.txt", "file2.txt"}
+
+
+@patch("onyx.server.documents.connector.get_default_file_store")
+def test_upload_invalid_zip_with_unzip_false_raises(
+    mock_get_store: MagicMock,
+) -> None:
+    """An invalid zip is rejected even when unzip=False (validation still runs)."""
+    mock_get_store.return_value = MagicMock()
+
+    bad_zip = _make_upload_file(b"not a zip", "bad.zip", "application/zip")
+
+    with pytest.raises(BadZipFile):
+        upload_files([bad_zip], FileOrigin.CONNECTOR, unzip=False)
+
+
+@patch("onyx.server.documents.connector.get_default_file_store")
+def test_upload_multiple_zips_rejected_when_unzip_false(
+    mock_get_store: MagicMock,
+) -> None:
+    """The seen_zip guard rejects a second zip even when unzip=False."""
+    mock_store = MagicMock()
+    mock_store.save_file.return_value = "zip-id"
+    mock_get_store.return_value = mock_store
+
+    zip_bytes = _create_test_zip()
+    zip1 = _make_upload_file(zip_bytes, "a.zip", "application/zip")
+    zip2 = _make_upload_file(zip_bytes, "b.zip", "application/zip")
+
+    with pytest.raises(Exception, match="Only one zip file"):
+        upload_files([zip1, zip2], FileOrigin.CONNECTOR, unzip=False)

backend/tests/unit/onyx/tools/tool_implementations/python/test_python_tool_upload_cache.py

@@ -0,0 +1,208 @@
+"""Unit tests for PythonTool file-upload caching.
+
+Verifies that PythonTool reuses code-interpreter file IDs across multiple
+run() calls within the same session instead of re-uploading identical content
+on every agent loop iteration.
+"""
+
+from unittest.mock import MagicMock
+from unittest.mock import patch
+
+from onyx.tools.models import ChatFile
+from onyx.tools.models import PythonToolOverrideKwargs
+from onyx.tools.tool_implementations.python.code_interpreter_client import (
+    StreamResultEvent,
+)
+from onyx.tools.tool_implementations.python.python_tool import PythonTool
+
+TOOL_MODULE = "onyx.tools.tool_implementations.python.python_tool"
+
+
+def _make_stream_result() -> StreamResultEvent:
+    return StreamResultEvent(
+        exit_code=0,
+        timed_out=False,
+        duration_ms=10,
+        files=[],
+    )
+
+
+def _make_tool() -> PythonTool:
+    emitter = MagicMock()
+    return PythonTool(tool_id=1, emitter=emitter)
+
+
+def _make_override(files: list[ChatFile]) -> PythonToolOverrideKwargs:
+    return PythonToolOverrideKwargs(chat_files=files)
+
+
+def _run_tool(tool: PythonTool, mock_client: MagicMock, files: list[ChatFile]) -> None:
+    """Call tool.run() with a mocked CodeInterpreterClient context manager."""
+    from onyx.server.query_and_chat.placement import Placement
+
+    mock_client.execute_streaming.return_value = iter([_make_stream_result()])
+
+    ctx = MagicMock()
+    ctx.__enter__ = MagicMock(return_value=mock_client)
+    ctx.__exit__ = MagicMock(return_value=False)
+
+    placement = Placement(turn_index=0, tab_index=0)
+    override = _make_override(files)
+
+    with patch(f"{TOOL_MODULE}.CodeInterpreterClient", return_value=ctx):
+        tool.run(placement=placement, override_kwargs=override, code="print('hi')")
+
+
+# ---------------------------------------------------------------------------
+# Cache hit: same content uploaded in a second call reuses the file_id
+# ---------------------------------------------------------------------------
+
+
+@patch(f"{TOOL_MODULE}.CODE_INTERPRETER_BASE_URL", "http://fake:8000")
+def test_same_file_uploaded_only_once_across_two_runs() -> None:
+    tool = _make_tool()
+    client = MagicMock()
+    client.upload_file.return_value = "file-id-abc"
+
+    pptx_content = b"fake pptx bytes"
+    files = [ChatFile(filename="report.pptx", content=pptx_content)]
+
+    _run_tool(tool, client, files)
+    _run_tool(tool, client, files)
+
+    # upload_file should only have been called once across both runs
+    client.upload_file.assert_called_once_with(pptx_content, "report.pptx")
+
+
+@patch(f"{TOOL_MODULE}.CODE_INTERPRETER_BASE_URL", "http://fake:8000")
+def test_cached_file_id_is_staged_on_second_run() -> None:
+    tool = _make_tool()
+    client = MagicMock()
+    client.upload_file.return_value = "file-id-abc"
+
+    files = [ChatFile(filename="data.pptx", content=b"content")]
+
+    _run_tool(tool, client, files)
+
+    # On the second run, execute_streaming should still receive the file
+    client.execute_streaming.return_value = iter([_make_stream_result()])
+    ctx = MagicMock()
+    ctx.__enter__ = MagicMock(return_value=client)
+    ctx.__exit__ = MagicMock(return_value=False)
+
+    from onyx.server.query_and_chat.placement import Placement
+
+    placement = Placement(turn_index=1, tab_index=0)
+    with patch(f"{TOOL_MODULE}.CodeInterpreterClient", return_value=ctx):
+        tool.run(
+            placement=placement,
+            override_kwargs=_make_override(files),
+            code="print('hi')",
+        )
+
+    # The second execute_streaming call should include the file
+    _, kwargs = client.execute_streaming.call_args
+    staged_files = kwargs.get("files") or []
+    assert any(f["file_id"] == "file-id-abc" for f in staged_files)
+
+
+# ---------------------------------------------------------------------------
+# Cache miss: different content triggers a new upload
+# ---------------------------------------------------------------------------
+
+
+@patch(f"{TOOL_MODULE}.CODE_INTERPRETER_BASE_URL", "http://fake:8000")
+def test_different_file_content_uploaded_separately() -> None:
+    tool = _make_tool()
+    client = MagicMock()
+    client.upload_file.side_effect = ["file-id-v1", "file-id-v2"]
+
+    file_v1 = ChatFile(filename="report.pptx", content=b"version 1")
+    file_v2 = ChatFile(filename="report.pptx", content=b"version 2")
+
+    _run_tool(tool, client, [file_v1])
+    _run_tool(tool, client, [file_v2])
+
+    assert client.upload_file.call_count == 2
+
+
+@patch(f"{TOOL_MODULE}.CODE_INTERPRETER_BASE_URL", "http://fake:8000")
+def test_multiple_distinct_files_each_uploaded_once() -> None:
+    tool = _make_tool()
+    client = MagicMock()
+    client.upload_file.side_effect = ["id-a", "id-b"]
+
+    files = [
+        ChatFile(filename="a.pptx", content=b"aaa"),
+        ChatFile(filename="b.xlsx", content=b"bbb"),
+    ]
+
+    _run_tool(tool, client, files)
+    _run_tool(tool, client, files)
+
+    # Two distinct files — each uploaded exactly once
+    assert client.upload_file.call_count == 2
+
+
+@patch(f"{TOOL_MODULE}.CODE_INTERPRETER_BASE_URL", "http://fake:8000")
+def test_same_content_different_filename_uploaded_separately() -> None:
+    # Identical bytes but different names must each get their own upload slot
+    # so both files appear under their respective paths in the workspace.
+    tool = _make_tool()
+    client = MagicMock()
+    client.upload_file.side_effect = ["id-v1", "id-v2"]
+
+    same_bytes = b"shared content"
+    files = [
+        ChatFile(filename="report_v1.csv", content=same_bytes),
+        ChatFile(filename="report_v2.csv", content=same_bytes),
+    ]
+
+    _run_tool(tool, client, files)
+
+    assert client.upload_file.call_count == 2
+
+
+# ---------------------------------------------------------------------------
+# No cross-instance sharing: a fresh PythonTool re-uploads everything
+# ---------------------------------------------------------------------------
+
+
+@patch(f"{TOOL_MODULE}.CODE_INTERPRETER_BASE_URL", "http://fake:8000")
+def test_new_tool_instance_re_uploads_file() -> None:
+    client = MagicMock()
+    client.upload_file.side_effect = ["id-session-1", "id-session-2"]
+
+    files = [ChatFile(filename="deck.pptx", content=b"slide data")]
+
+    tool_session_1 = _make_tool()
+    _run_tool(tool_session_1, client, files)
+
+    tool_session_2 = _make_tool()
+    _run_tool(tool_session_2, client, files)
+
+    # Different instances — each uploads independently
+    assert client.upload_file.call_count == 2
+
+
+# ---------------------------------------------------------------------------
+# Upload failure: failed upload is not cached, retried next run
+# ---------------------------------------------------------------------------
+
+
+@patch(f"{TOOL_MODULE}.CODE_INTERPRETER_BASE_URL", "http://fake:8000")
+def test_upload_failure_not_cached() -> None:
+    tool = _make_tool()
+    client = MagicMock()
+    # First call raises, second succeeds
+    client.upload_file.side_effect = [Exception("network error"), "file-id-ok"]
+
+    files = [ChatFile(filename="slides.pptx", content=b"data")]
+
+    # First run — upload fails, file is skipped but not cached
+    _run_tool(tool, client, files)
+
+    # Second run — should attempt upload again
+    _run_tool(tool, client, files)
+
+    assert client.upload_file.call_count == 2

cubic.yaml

@@ -0,0 +1,93 @@
+# yaml-language-server: $schema=https://cubic.dev/schema/cubic-repository-config.schema.json
+version: 1
+
+reviews:
+  enabled: true
+  sensitivity: medium
+  incremental_commits: true
+  check_drafts: false
+
+  custom_instructions: |
+    Use explicit type annotations for variables to enhance code clarity,
+    especially when moving type hints around in the code.
+
+    Use `contributing_guides/best_practices.md` as core review context.
+    Prefer consistency with existing patterns, fix issues in code you touch,
+    avoid tacking new features onto muddy interfaces, fail loudly instead of
+    silently swallowing errors, keep code strictly typed, preserve clear state
+    boundaries, remove duplicate or dead logic, break up overly long functions,
+    avoid hidden import-time side effects, respect module boundaries, and favor
+    correctness-by-construction over relying on callers to use an API correctly.
+
+    Reference these files for additional context:
+    - `contributing_guides/best_practices.md` — Best practices for contributing to the codebase
+    - `CLAUDE.md` — Project instructions and coding standards
+    - `backend/alembic/README.md` — Migration guidance, including multi-tenant migration behavior
+    - `deployment/helm/charts/onyx/values-lite.yaml` — Lite deployment Helm values and service assumptions
+    - `deployment/docker_compose/docker-compose.onyx-lite.yml` — Lite deployment Docker Compose overlay and disabled service behavior
+
+  ignore:
+    files:
+      - greptile.json
+      - cubic.yaml
+
+  custom_rules:
+    - name: TODO format
+      description: >
+        Whenever a TODO is added, there must always be an associated name or
+        ticket in the style of TODO(name): ... or TODO(1234): ...
+
+    - name: Frontend standards
+      description: >
+        For frontend changes, enforce all standards described in the
+        web/AGENTS.md file.
+      include:
+        - web/**
+        - desktop/**
+
+    - name: No debugging code
+      description: >
+        Remove temporary debugging code before merging to production,
+        especially tenant-specific debugging logs.
+
+    - name: No hardcoded booleans
+      description: >
+        When hardcoding a boolean variable to a constant value, remove the
+        variable entirely and clean up all places where it's used rather than
+        just setting it to a constant.
+
+    - name: Multi-tenant awareness
+      description: >
+        Code changes must consider both multi-tenant and single-tenant
+        deployments. In multi-tenant mode, preserve tenant isolation, ensure
+        tenant context is propagated correctly, and avoid assumptions that only
+        hold for a single shared schema or globally shared state. In
+        single-tenant mode, avoid introducing unnecessary tenant-specific
+        requirements or cloud-only control-plane dependencies.
+
+    - name: Onyx lite compatibility
+      description: >
+        Code changes must consider both regular Onyx deployments and Onyx lite
+        deployments. Lite deployments disable the vector DB, Redis, model
+        servers, and background workers by default, use PostgreSQL-backed
+        cache/auth/file storage, and rely on the API server to handle
+        background work. Do not assume those services are available unless the
+        code path is explicitly limited to full deployments.
+
+    - name: OnyxError over HTTPException
+      description: >
+        Never raise HTTPException directly in business code. Use
+        `raise OnyxError(OnyxErrorCode.XXX, "message")` from
+        `onyx.error_handling.exceptions`. A global FastAPI exception handler
+        converts OnyxError into structured JSON responses with
+        {"error_code": "...", "detail": "..."}. Error codes are defined in
+        `onyx.error_handling.error_codes.OnyxErrorCode`. For upstream errors
+        with dynamic HTTP status codes, use `status_code_override`:
+        `raise OnyxError(OnyxErrorCode.BAD_GATEWAY, detail, status_code_override=upstream_status)`.
+      include:
+        - backend/**/*.py
+
+issues:
+  fix_with_cubic_buttons: true
+  pr_comment_fixes: true
+  fix_commits_to_pr: true

deployment/docker_compose/docker-compose.dev.yml

@@ -12,6 +12,11 @@ services:
   api_server:
     ports:
       - "8080:8080"
+    deploy:
+      resources:
+        limits:
+          cpus: "${API_SERVER_CPU_LIMIT:-0}"
+          memory: "${API_SERVER_MEM_LIMIT:-0}"
 
   # Uncomment the block below to enable the MCP server for Onyx.
   # mcp_server:

deployment/docker_compose/docker-compose.multitenant-dev.yml

@@ -489,20 +489,18 @@ services:
       - "${HOST_PORT_80:-80}:80"
       - "${HOST_PORT:-3000}:80" # allow for localhost:3000 usage, since that is the norm
     volumes:
-      - ../data/nginx:/etc/nginx/conf.d
+      - ../data/nginx:/nginx-templates:ro
     logging:
       driver: json-file
       options:
         max-size: "50m"
         max-file: "6"
-    # The specified script waits for the api_server to start up.
-    # Without this we've seen issues where nginx shows no error logs but
-    # does not recieve any traffic
-    # NOTE: we have to use dos2unix to remove Carriage Return chars from the file
-    # in order to make this work on both Unix-like systems and windows
     command: >
-      /bin/sh -c "dos2unix /etc/nginx/conf.d/run-nginx.sh
-      && /etc/nginx/conf.d/run-nginx.sh app.conf.template"
+      /bin/sh -c "rm -f /etc/nginx/conf.d/default.conf
+      && cp -a /nginx-templates/. /etc/nginx/conf.d/
+      && sed 's/\r$//' /etc/nginx/conf.d/run-nginx.sh > /tmp/run-nginx.sh
+      && chmod +x /tmp/run-nginx.sh
+      && /tmp/run-nginx.sh app.conf.template"
 
   minio:
     image: minio/minio:RELEASE.2025-07-23T15-54-02Z-cpuv1

deployment/docker_compose/docker-compose.prod-cloud.yml

@@ -290,25 +290,20 @@ services:
       - "80:80"
       - "443:443"
     volumes:
-      - ../data/nginx:/etc/nginx/conf.d
+      - ../data/nginx:/nginx-templates:ro
       - ../data/certbot/conf:/etc/letsencrypt
       - ../data/certbot/www:/var/www/certbot
-    # sleep a little bit to allow the web_server / api_server to start up.
-    # Without this we've seen issues where nginx shows no error logs but
-    # does not recieve any traffic
     logging:
       driver: json-file
       options:
         max-size: "50m"
         max-file: "6"
-    # The specified script waits for the api_server to start up.
-    # Without this we've seen issues where nginx shows no error logs but
-    # does not recieve any traffic
-    # NOTE: we have to use dos2unix to remove Carriage Return chars from the file
-    # in order to make this work on both Unix-like systems and windows
     command: >
-      /bin/sh -c "dos2unix /etc/nginx/conf.d/run-nginx.sh
-      && /etc/nginx/conf.d/run-nginx.sh app.conf.template.prod"
+      /bin/sh -c "rm -f /etc/nginx/conf.d/default.conf
+      && cp -a /nginx-templates/. /etc/nginx/conf.d/
+      && sed 's/\r$//' /etc/nginx/conf.d/run-nginx.sh > /tmp/run-nginx.sh
+      && chmod +x /tmp/run-nginx.sh
+      && /tmp/run-nginx.sh app.conf.template.prod"
     env_file:
       - .env.nginx
     environment:

deployment/docker_compose/docker-compose.prod-no-letsencrypt.yml

@@ -314,21 +314,19 @@ services:
       - "80:80"
       - "443:443"
     volumes:
-      - ../data/nginx:/etc/nginx/conf.d
+      - ../data/nginx:/nginx-templates:ro
       - ../data/sslcerts:/etc/nginx/sslcerts
     logging:
       driver: json-file
       options:
         max-size: "50m"
         max-file: "6"
-    # The specified script waits for the api_server to start up.
-    # Without this we've seen issues where nginx shows no error logs but
-    # does not recieve any traffic
-    # NOTE: we have to use dos2unix to remove Carriage Return chars from the file
-    # in order to make this work on both Unix-like systems and windows
     command: >
-      /bin/sh -c "dos2unix /etc/nginx/conf.d/run-nginx.sh
-      && /etc/nginx/conf.d/run-nginx.sh app.conf.template.prod.no-letsencrypt"
+      /bin/sh -c "rm -f /etc/nginx/conf.d/default.conf
+      && cp -a /nginx-templates/. /etc/nginx/conf.d/
+      && sed 's/\r$//' /etc/nginx/conf.d/run-nginx.sh > /tmp/run-nginx.sh
+      && chmod +x /tmp/run-nginx.sh
+      && /tmp/run-nginx.sh app.conf.template.prod.no-letsencrypt"
     env_file:
       - .env.nginx
     environment:

deployment/docker_compose/docker-compose.prod.yml

@@ -333,25 +333,20 @@ services:
       - "80:80"
       - "443:443"
     volumes:
-      - ../data/nginx:/etc/nginx/conf.d
+      - ../data/nginx:/nginx-templates:ro
       - ../data/certbot/conf:/etc/letsencrypt
       - ../data/certbot/www:/var/www/certbot
-    # sleep a little bit to allow the web_server / api_server to start up.
-    # Without this we've seen issues where nginx shows no error logs but
-    # does not recieve any traffic
     logging:
       driver: json-file
       options:
         max-size: "50m"
         max-file: "6"
-    # The specified script waits for the api_server to start up.
-    # Without this we've seen issues where nginx shows no error logs but
-    # does not recieve any traffic
-    # NOTE: we have to use dos2unix to remove Carriage Return chars from the file
-    # in order to make this work on both Unix-like systems and windows
     command: >
-      /bin/sh -c "dos2unix /etc/nginx/conf.d/run-nginx.sh
-      && /etc/nginx/conf.d/run-nginx.sh app.conf.template.prod"
+      /bin/sh -c "rm -f /etc/nginx/conf.d/default.conf
+      && cp -a /nginx-templates/. /etc/nginx/conf.d/
+      && sed 's/\r$//' /etc/nginx/conf.d/run-nginx.sh > /tmp/run-nginx.sh
+      && chmod +x /tmp/run-nginx.sh
+      && /tmp/run-nginx.sh app.conf.template.prod"
     env_file:
       - .env.nginx
     environment:

deployment/docker_compose/docker-compose.search-testing.yml

@@ -202,20 +202,18 @@ services:
     ports:
       - "${NGINX_PORT:-3000}:80" # allow for localhost:3000 usage, since that is the norm
     volumes:
-      - ../data/nginx:/etc/nginx/conf.d
+      - ../data/nginx:/nginx-templates:ro
     logging:
       driver: json-file
       options:
         max-size: "50m"
         max-file: "6"
-    # The specified script waits for the api_server to start up.
-    # Without this we've seen issues where nginx shows no error logs but
-    # does not recieve any traffic
-    # NOTE: we have to use dos2unix to remove Carriage Return chars from the file
-    # in order to make this work on both Unix-like systems and windows
     command: >
-      /bin/sh -c "dos2unix /etc/nginx/conf.d/run-nginx.sh
-      && /etc/nginx/conf.d/run-nginx.sh app.conf.template"
+      /bin/sh -c "rm -f /etc/nginx/conf.d/default.conf
+      && cp -a /nginx-templates/. /etc/nginx/conf.d/
+      && sed 's/\r$//' /etc/nginx/conf.d/run-nginx.sh > /tmp/run-nginx.sh
+      && chmod +x /tmp/run-nginx.sh
+      && /tmp/run-nginx.sh app.conf.template"
 
   minio:
     image: minio/minio:RELEASE.2025-07-23T15-54-02Z-cpuv1

deployment/docker_compose/docker-compose.yml

@@ -477,7 +477,10 @@ services:
       - "${HOST_PORT_80:-80}:80"
       - "${HOST_PORT:-3000}:80" # allow for localhost:3000 usage, since that is the norm
     volumes:
-      - ../data/nginx:/etc/nginx/conf.d
+      # Mount templates read-only; the startup command copies them into
+      # the writable /etc/nginx/conf.d/ inside the container.  This avoids
+      # "Permission denied" errors on Windows Docker bind mounts.
+      - ../data/nginx:/nginx-templates:ro
     # PRODUCTION: Add SSL certificate volumes for HTTPS support:
     #   - ../data/certbot/conf:/etc/letsencrypt
     #   - ../data/certbot/www:/var/www/certbot
@@ -489,12 +492,13 @@ services:
     # The specified script waits for the api_server to start up.
     # Without this we've seen issues where nginx shows no error logs but
     # does not receive any traffic
-    # NOTE: we have to use dos2unix to remove Carriage Return chars from the file
-    # in order to make this work on both Unix-like systems and windows
     # PRODUCTION: Change to app.conf.template.prod for production nginx config
     command: >
-      /bin/sh -c "dos2unix /etc/nginx/conf.d/run-nginx.sh
-      && /etc/nginx/conf.d/run-nginx.sh app.conf.template"
+      /bin/sh -c "rm -f /etc/nginx/conf.d/default.conf
+      && cp -a /nginx-templates/. /etc/nginx/conf.d/
+      && sed 's/\r$//' /etc/nginx/conf.d/run-nginx.sh > /tmp/run-nginx.sh
+      && chmod +x /tmp/run-nginx.sh
+      && /tmp/run-nginx.sh app.conf.template"
 
   cache:
     image: redis:7.4-alpine

deployment/docker_compose/install.sh

@@ -96,8 +96,8 @@ fi
 
 # When --lite is passed as a flag, lower resource thresholds early (before the
 # resource check). When lite is chosen interactively, the thresholds are adjusted
-# inside the new-deployment flow, after the resource check has already passed
-# with the standard thresholds — which is the safer direction.
+# after the resource check has already passed with the standard thresholds —
+# which is the safer direction.
 if [[ "$LITE_MODE" = true ]]; then
     EXPECTED_DOCKER_RAM_GB=4
     EXPECTED_DISK_GB=16
@@ -110,9 +110,6 @@ LITE_COMPOSE_FILE="docker-compose.onyx-lite.yml"
 # Build the -f flags for docker compose.
 # Pass "true" as $1 to auto-detect a previously-downloaded lite overlay
 # (used by shutdown/delete-data so users don't need to remember --lite).
-# Without the argument, the lite overlay is only included when --lite was
-# explicitly passed — preventing install/start from silently staying in
-# lite mode just because the file exists on disk from a prior run.
 compose_file_args() {
     local auto_detect="${1:-false}"
     local args="-f docker-compose.yml"
@@ -177,7 +174,7 @@ ensure_file() {
 
 # --- Interactive prompt helpers ---
 is_interactive() {
-    [[ "$NO_PROMPT" = false ]] && [[ -t 0 ]]
+    [[ "$NO_PROMPT" = false ]]
 }
 
 prompt_or_default() {
@@ -207,6 +204,16 @@ prompt_yn_or_default() {
     fi
 }
 
+confirm_action() {
+    local description="$1"
+    prompt_yn_or_default "Install ${description}? (Y/n) [default: Y] " "Y"
+    if [[ "$REPLY" =~ ^[Nn] ]]; then
+        print_warning "Skipping: ${description}"
+        return 1
+    fi
+    return 0
+}
+
 # Colors for output
 RED='\033[0;31m'
 GREEN='\033[0;32m'
@@ -395,6 +402,11 @@ fi
 
 if ! command -v docker &> /dev/null; then
     if [[ "$OSTYPE" == "linux-gnu"* ]] || [[ -n "${WSL_DISTRO_NAME:-}" ]]; then
+        print_info "Docker is required but not installed."
+        if ! confirm_action "Docker Engine"; then
+            print_error "Docker is required to run Onyx."
+            exit 1
+        fi
         install_docker_linux
         if ! command -v docker &> /dev/null; then
             print_error "Docker installation failed."
@@ -411,7 +423,11 @@ if command -v docker &> /dev/null \
     && ! command -v docker-compose &> /dev/null \
     && { [[ "$OSTYPE" == "linux-gnu"* ]] || [[ -n "${WSL_DISTRO_NAME:-}" ]]; }; then
 
-    print_info "Docker Compose not found — installing plugin..."
+    print_info "Docker Compose is required but not installed."
+    if ! confirm_action "Docker Compose plugin"; then
+        print_error "Docker Compose is required to run Onyx."
+        exit 1
+    fi
     COMPOSE_ARCH="$(uname -m)"
     COMPOSE_URL="https://github.com/docker/compose/releases/latest/download/docker-compose-linux-${COMPOSE_ARCH}"
     COMPOSE_DIR="/usr/local/lib/docker/cli-plugins"
@@ -562,10 +578,31 @@ version_compare() {
 
 # Check Docker daemon
 if ! docker info &> /dev/null; then
-    print_error "Docker daemon is not running. Please start Docker."
-    exit 1
+    if [[ "$OSTYPE" == "darwin"* ]]; then
+        print_info "Docker daemon is not running. Starting Docker Desktop..."
+        open -a Docker
+        # Wait up to 120 seconds for Docker to be ready
+        DOCKER_WAIT=0
+        DOCKER_MAX_WAIT=120
+        while ! docker info &> /dev/null; do
+            if [ $DOCKER_WAIT -ge $DOCKER_MAX_WAIT ]; then
+                print_error "Docker Desktop did not start within ${DOCKER_MAX_WAIT} seconds."
+                print_info "Please start Docker Desktop manually and re-run this script."
+                exit 1
+            fi
+            printf "\r\033[KWaiting for Docker Desktop to start... (%ds)" "$DOCKER_WAIT"
+            sleep 2
+            DOCKER_WAIT=$((DOCKER_WAIT + 2))
+        done
+        echo ""
+        print_success "Docker Desktop is now running"
+    else
+        print_error "Docker daemon is not running. Please start Docker."
+        exit 1
+    fi
+else
+    print_success "Docker daemon is running"
 fi
-print_success "Docker daemon is running"
 
 # Check Docker resources
 print_step "Verifying Docker resources"
@@ -705,25 +742,48 @@ if [ "$COMPOSE_VERSION" != "dev" ] && version_compare "$COMPOSE_VERSION" "2.24.0
     print_info "Proceeding with installation despite Docker Compose version compatibility issues..."
 fi
 
-# Handle lite overlay: ensure it if --lite, clean up stale copies otherwise
+# Ask for deployment mode (standard vs lite) unless already set via --lite flag
+if [[ "$LITE_MODE" = false ]]; then
+    print_info "Which deployment mode would you like?"
+    echo ""
+    echo "  1) Lite      - Minimal deployment (no Vespa, Redis, or model servers)"
+    echo "                  LLM chat, tools, file uploads, and Projects still work"
+    echo "  2) Standard  - Full deployment with search, connectors, and RAG"
+    echo ""
+    prompt_or_default "Choose a mode (1 or 2) [default: 1]: " "1"
+    echo ""
+
+    case "$REPLY" in
+        2)
+            print_info "Selected: Standard mode"
+            ;;
+        *)
+            LITE_MODE=true
+            print_info "Selected: Lite mode"
+            ;;
+    esac
+else
+    print_info "Deployment mode: Lite (set via --lite flag)"
+fi
+
+if [[ "$LITE_MODE" = true ]] && [[ "$INCLUDE_CRAFT" = true ]]; then
+    print_error "--include-craft cannot be used with Lite mode."
+    print_info "Craft requires services (Vespa, Redis, background workers) that lite mode disables."
+    exit 1
+fi
+
+if [[ "$LITE_MODE" = true ]]; then
+    EXPECTED_DOCKER_RAM_GB=4
+    EXPECTED_DISK_GB=16
+fi
+
+# Handle lite overlay file based on selected mode
 if [[ "$LITE_MODE" = true ]]; then
     ensure_file "${INSTALL_ROOT}/deployment/${LITE_COMPOSE_FILE}" \
         "${GITHUB_RAW_URL}/${LITE_COMPOSE_FILE}" "${LITE_COMPOSE_FILE}" || exit 1
 elif [[ -f "${INSTALL_ROOT}/deployment/${LITE_COMPOSE_FILE}" ]]; then
-    if [[ -f "${INSTALL_ROOT}/deployment/.env" ]]; then
-        print_warning "Existing lite overlay found but --lite was not passed."
-        prompt_yn_or_default "Remove lite overlay and switch to standard mode? (y/N): " "n"
-        if [[ ! $REPLY =~ ^[Yy]$ ]]; then
-            print_info "Keeping existing lite overlay. Pass --lite to keep using lite mode."
-            LITE_MODE=true
-        else
-            rm -f "${INSTALL_ROOT}/deployment/${LITE_COMPOSE_FILE}"
-            print_info "Removed lite overlay (switching to standard mode)"
-        fi
-    else
-        rm -f "${INSTALL_ROOT}/deployment/${LITE_COMPOSE_FILE}"
-        print_info "Removed previous lite overlay (switching to standard mode)"
-    fi
+    rm -f "${INSTALL_ROOT}/deployment/${LITE_COMPOSE_FILE}"
+    print_info "Removed previous lite overlay (switching to standard mode)"
 fi
 
 ensure_file "${INSTALL_ROOT}/deployment/env.template" \
@@ -745,6 +805,7 @@ print_success "All configuration files ready"
 # Set up deployment configuration
 print_step "Setting up deployment configs"
 ENV_FILE="${INSTALL_ROOT}/deployment/.env"
+ENV_TEMPLATE="${INSTALL_ROOT}/deployment/env.template"
 # Check if services are already running
 if [ -d "${INSTALL_ROOT}/deployment" ] && [ -f "${INSTALL_ROOT}/deployment/docker-compose.yml" ]; then
     # Determine compose command
@@ -785,22 +846,22 @@ if [ -f "$ENV_FILE" ]; then
     if [ "$REPLY" = "update" ]; then
         print_info "Update selected. Which tag would you like to deploy?"
         echo ""
-        echo "• Press Enter for latest (recommended)"
+        echo "• Press Enter for edge (recommended)"
         echo "• Type a specific tag (e.g., v0.1.0)"
         echo ""
         if [ "$INCLUDE_CRAFT" = true ]; then
             prompt_or_default "Enter tag [default: craft-latest]: " "craft-latest"
             VERSION="$REPLY"
         else
-            prompt_or_default "Enter tag [default: latest]: " "latest"
+            prompt_or_default "Enter tag [default: edge]: " "edge"
             VERSION="$REPLY"
         fi
         echo ""
 
         if [ "$INCLUDE_CRAFT" = true ] && [ "$VERSION" = "craft-latest" ]; then
             print_info "Selected: craft-latest (Craft enabled)"
-        elif [ "$VERSION" = "latest" ]; then
-            print_info "Selected: Latest version"
+        elif [ "$VERSION" = "edge" ]; then
+            print_info "Selected: edge (latest nightly)"
         else
             print_info "Selected: $VERSION"
         fi
@@ -852,45 +913,6 @@ else
     print_info "No existing .env file found. Setting up new deployment..."
     echo ""
 
-    # Ask for deployment mode (standard vs lite) unless already set via --lite flag
-    if [[ "$LITE_MODE" = false ]]; then
-        print_info "Which deployment mode would you like?"
-        echo ""
-        echo "  1) Standard  - Full deployment with search, connectors, and RAG"
-        echo "  2) Lite      - Minimal deployment (no Vespa, Redis, or model servers)"
-        echo "                  LLM chat, tools, file uploads, and Projects still work"
-        echo ""
-        prompt_or_default "Choose a mode (1 or 2) [default: 1]: " "1"
-        echo ""
-
-        case "$REPLY" in
-            2)
-                LITE_MODE=true
-                print_info "Selected: Lite mode"
-                ensure_file "${INSTALL_ROOT}/deployment/${LITE_COMPOSE_FILE}" \
-                    "${GITHUB_RAW_URL}/${LITE_COMPOSE_FILE}" "${LITE_COMPOSE_FILE}" || exit 1
-                ;;
-            *)
-                print_info "Selected: Standard mode"
-                ;;
-        esac
-    else
-        print_info "Deployment mode: Lite (set via --lite flag)"
-    fi
-
-    # Validate lite + craft combination (could now be set interactively)
-    if [[ "$LITE_MODE" = true ]] && [[ "$INCLUDE_CRAFT" = true ]]; then
-        print_error "--include-craft cannot be used with Lite mode."
-        print_info "Craft requires services (Vespa, Redis, background workers) that lite mode disables."
-        exit 1
-    fi
-
-    # Adjust resource expectations for lite mode
-    if [[ "$LITE_MODE" = true ]]; then
-        EXPECTED_DOCKER_RAM_GB=4
-        EXPECTED_DISK_GB=16
-    fi
-
     # Ask for version
     print_info "Which tag would you like to deploy?"
     echo ""
@@ -901,18 +923,18 @@ else
         prompt_or_default "Enter tag [default: craft-latest]: " "craft-latest"
         VERSION="$REPLY"
     else
-        echo "• Press Enter for latest (recommended)"
+        echo "• Press Enter for edge (recommended)"
         echo "• Type a specific tag (e.g., v0.1.0)"
         echo ""
-        prompt_or_default "Enter tag [default: latest]: " "latest"
+        prompt_or_default "Enter tag [default: edge]: " "edge"
         VERSION="$REPLY"
     fi
     echo ""
 
     if [ "$INCLUDE_CRAFT" = true ] && [ "$VERSION" = "craft-latest" ]; then
         print_info "Selected: craft-latest (Craft enabled)"
-    elif [ "$VERSION" = "latest" ]; then
-        print_info "Selected: Latest tag"
+    elif [ "$VERSION" = "edge" ]; then
+        print_info "Selected: edge (latest nightly)"
     else
         print_info "Selected: $VERSION"
     fi
@@ -1070,20 +1092,39 @@ fi
 export HOST_PORT=$AVAILABLE_PORT
 print_success "Using port $AVAILABLE_PORT for nginx"
 
-# Determine if we're using the latest tag or a craft tag (both should force pull)
+# Determine if we're using a floating tag (edge, latest, craft-*) that should force pull
 # Read IMAGE_TAG from .env file and remove any quotes or whitespace
 CURRENT_IMAGE_TAG=$(grep "^IMAGE_TAG=" "$ENV_FILE" | head -1 | cut -d'=' -f2 | tr -d ' "'"'"'')
-if [ "$CURRENT_IMAGE_TAG" = "latest" ] || [[ "$CURRENT_IMAGE_TAG" == craft-* ]]; then
+if [ "$CURRENT_IMAGE_TAG" = "edge" ] || [ "$CURRENT_IMAGE_TAG" = "latest" ] || [[ "$CURRENT_IMAGE_TAG" == craft-* ]]; then
     USE_LATEST=true
     if [[ "$CURRENT_IMAGE_TAG" == craft-* ]]; then
         print_info "Using craft tag '$CURRENT_IMAGE_TAG' - will force pull and recreate containers"
     else
-        print_info "Using 'latest' tag - will force pull and recreate containers"
+        print_info "Using '$CURRENT_IMAGE_TAG' tag - will force pull and recreate containers"
     fi
 else
     USE_LATEST=false
 fi
 
+# For pinned version tags, re-download config files from that tag so the
+# compose file matches the images being pulled (the initial download used main).
+if [[ "$USE_LATEST" = false ]] && [[ "$USE_LOCAL_FILES" = false ]]; then
+    PINNED_BASE="https://raw.githubusercontent.com/onyx-dot-app/onyx/${CURRENT_IMAGE_TAG}/deployment"
+    print_info "Fetching config files matching tag ${CURRENT_IMAGE_TAG}..."
+    if download_file "${PINNED_BASE}/docker_compose/docker-compose.yml" "${INSTALL_ROOT}/deployment/docker-compose.yml" 2>/dev/null; then
+        download_file "${PINNED_BASE}/data/nginx/app.conf.template" "${INSTALL_ROOT}/data/nginx/app.conf.template" 2>/dev/null || true
+        download_file "${PINNED_BASE}/data/nginx/run-nginx.sh" "${INSTALL_ROOT}/data/nginx/run-nginx.sh" 2>/dev/null || true
+        chmod +x "${INSTALL_ROOT}/data/nginx/run-nginx.sh"
+        if [[ "$LITE_MODE" = true ]]; then
+            download_file "${PINNED_BASE}/docker_compose/${LITE_COMPOSE_FILE}" \
+                "${INSTALL_ROOT}/deployment/${LITE_COMPOSE_FILE}" 2>/dev/null || true
+        fi
+        print_success "Config files updated to match ${CURRENT_IMAGE_TAG}"
+    else
+        print_warning "Tag ${CURRENT_IMAGE_TAG} not found on GitHub — using main branch configs"
+    fi
+fi
+
 # Pull Docker images with reduced output
 print_step "Pulling Docker images"
 print_info "This may take several minutes depending on your internet connection..."

deployment/helm/README.md

@@ -19,12 +19,13 @@
 
 ## Output template to file and inspect
 * cd charts/onyx
-* helm template test-output . > test-output.yaml
+* helm template test-output . --set auth.opensearch.values.opensearch_admin_password='StrongPassword123!' > test-output.yaml
 
 ## Test the entire cluster manually
 * cd charts/onyx
-* helm install onyx . -n onyx --set postgresql.primary.persistence.enabled=false
+* helm install onyx . -n onyx --set postgresql.primary.persistence.enabled=false --set auth.opensearch.values.opensearch_admin_password='StrongPassword123!'
   * the postgres flag is to keep the storage ephemeral for testing. You probably don't want to set that in prod.
+  * the OpenSearch admin password must be set on first install unless you are supplying `auth.opensearch.existingSecret`.
   * no flag for ephemeral vespa storage yet, might be good for testing
 * kubectl -n onyx port-forward service/onyx-nginx 8080:80
   * this will forward the local port 8080 to the installed chart for you to run tests, etc.

deployment/helm/charts/onyx/Chart.yaml

@@ -5,7 +5,7 @@ home: https://www.onyx.app/
 sources:
   - "https://github.com/onyx-dot-app/onyx"
 type: application
-version: 0.4.35
+version: 0.4.36
 appVersion: latest
 annotations:
   category: Productivity

deployment/helm/charts/onyx/ci/ct-values.yaml

@@ -1,6 +1,9 @@
 # Values for chart-testing (ct lint/install)
 # This file is automatically used by ct when running lint and install commands
 auth:
+  opensearch:
+    values:
+      opensearch_admin_password: "placeholder-OpenSearch1!"
   userauth:
     values:
       user_auth_secret: "placeholder-for-ci-testing"

deployment/helm/charts/onyx/values.yaml

@@ -1177,12 +1177,17 @@ auth:
     # Secrets values IF existingSecret is empty. Key here must match the value
     # in secretKeys to be used. Values will be base64 encoded in the k8s
     # cluster.
+    # For the bundled OpenSearch chart, the admin password is consumed during
+    # initial cluster setup. Changing this value later will update Onyx's
+    # client credentials, but will not rotate the OpenSearch admin password.
+    # Set this before first install or use existingSecret to preserve the
+    # current secret on upgrade.
     # Password must meet OpenSearch complexity requirements:
     # min 8 chars, uppercase, lowercase, digit, and special character.
-    # CHANGE THIS FOR PRODUCTION.
+    # Required when auth.opensearch.enabled=true and no existing secret exists.
     values:
       opensearch_admin_username: "admin"
-      opensearch_admin_password: "OnyxDev1!"
+      opensearch_admin_password: ""
   userauth:
     # -- Used for password reset / verification tokens and OAuth/OIDC state signing.
     # Disabled by default to preserve upgrade compatibility for existing Helm customers.

deployment/terraform/modules/aws/README.md

@@ -127,6 +127,7 @@ Inputs (common):
 - `name` (default `onyx`), `region` (default `us-west-2`), `tags`
 - `postgres_username`, `postgres_password`
 - `create_vpc` (default true) or existing VPC details and `s3_vpc_endpoint_id`
+- WAF controls such as `waf_allowed_ip_cidrs`, `waf_common_rule_set_count_rules`, rate limits, geo restrictions, and logging retention
 
 ### `vpc`
 - Builds a VPC sized for EKS with multiple private and public subnets

deployment/terraform/modules/aws/onyx/main.tf

@@ -88,6 +88,8 @@ module "waf" {
   tags = local.merged_tags
 
   # WAF configuration with sensible defaults
+  allowed_ip_cidrs                      = var.waf_allowed_ip_cidrs
+  common_rule_set_count_rules           = var.waf_common_rule_set_count_rules
   rate_limit_requests_per_5_minutes     = var.waf_rate_limit_requests_per_5_minutes
   api_rate_limit_requests_per_5_minutes = var.waf_api_rate_limit_requests_per_5_minutes
   geo_restriction_countries             = var.waf_geo_restriction_countries