feat(ods): Rerun run-ci workflow (#9501)

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

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

@@ -9,6 +9,7 @@ from onyx.configs.app_configs import AUTO_LLM_UPDATE_INTERVAL_SECONDS
 from onyx.configs.app_configs import DISABLE_VECTOR_DB
 from onyx.configs.app_configs import ENABLE_OPENSEARCH_INDEXING_FOR_ONYX
 from onyx.configs.app_configs import ENTERPRISE_EDITION_ENABLED
+from onyx.configs.app_configs import HOOK_ENABLED
 from onyx.configs.app_configs import SCHEDULED_EVAL_DATASET_NAMES
 from onyx.configs.constants import ONYX_CLOUD_CELERY_TASK_PREFIX
 from onyx.configs.constants import OnyxCeleryPriority
@@ -361,6 +362,19 @@ if not MULTI_TENANT:
 
     tasks_to_schedule.extend(beat_task_templates)
 
+if not MULTI_TENANT and HOOK_ENABLED:
+    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/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"
@@ -776,9 +787,6 @@ MINI_CHUNK_SIZE = 150
 # This is the number of regular chunks per large chunk
 LARGE_CHUNK_RATIO = 4
 
-# The number of chunks in an indexing batch
-CHUNKS_PER_BATCH = 1000
-
 # Include the document level metadata in each chunk. If the metadata is too long, then it is thrown out
 # We don't want the metadata to overwhelm the actual contents of the chunk
 SKIP_METADATA_IN_CHUNK = os.environ.get("SKIP_METADATA_IN_CHUNK", "").lower() == "true"

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/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/document_index/disabled.py

@@ -5,7 +5,6 @@ accidentally reaches the vector DB layer will fail loudly instead of timing
 out against a nonexistent Vespa/OpenSearch instance.
 """
 
-from collections.abc import Iterable
 from typing import Any
 
 from onyx.context.search.models import IndexFilters
@@ -67,7 +66,7 @@ class DisabledDocumentIndex(DocumentIndex):
     # ------------------------------------------------------------------
     def index(
         self,
-        chunks: Iterable[DocMetadataAwareIndexChunk],  # noqa: ARG002
+        chunks: list[DocMetadataAwareIndexChunk],  # noqa: ARG002
         index_batch_params: IndexBatchParams,  # noqa: ARG002
     ) -> set[DocumentInsertionRecord]:
         raise RuntimeError(VECTOR_DB_DISABLED_ERROR)

backend/onyx/document_index/interfaces.py

@@ -1,5 +1,4 @@
 import abc
-from collections.abc import Iterable
 from dataclasses import dataclass
 from datetime import datetime
 from typing import Any
@@ -207,7 +206,7 @@ class Indexable(abc.ABC):
     @abc.abstractmethod
     def index(
         self,
-        chunks: Iterable[DocMetadataAwareIndexChunk],
+        chunks: list[DocMetadataAwareIndexChunk],
         index_batch_params: IndexBatchParams,
     ) -> set[DocumentInsertionRecord]:
         """
@@ -227,8 +226,8 @@ class Indexable(abc.ABC):
         it is done automatically outside of this code.
 
         Parameters:
-        - chunks: Document chunks with all of the information needed for
-                indexing to the document index.
+        - chunks: Document chunks with all of the information needed for indexing to the document
+                index.
         - tenant_id: The tenant id of the user whose chunks are being indexed
         - large_chunks_enabled: Whether large chunks are enabled
 

backend/onyx/document_index/interfaces_new.py

@@ -1,5 +1,4 @@
 import abc
-from collections.abc import Iterable
 from typing import Self
 
 from pydantic import BaseModel
@@ -210,10 +209,10 @@ class Indexable(abc.ABC):
     @abc.abstractmethod
     def index(
         self,
-        chunks: Iterable[DocMetadataAwareIndexChunk],
+        chunks: list[DocMetadataAwareIndexChunk],
         indexing_metadata: IndexingMetadata,
     ) -> list[DocumentInsertionRecord]:
-        """Indexes an iterable of document chunks into the document index.
+        """Indexes a list of document chunks into the document index.
 
         This is often a batch operation including chunks from multiple
         documents.

backend/onyx/document_index/opensearch/client.py

@@ -18,6 +18,7 @@ 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 +57,8 @@ 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 +834,13 @@ 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 +888,7 @@ 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 +898,10 @@ 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,

backend/onyx/document_index/opensearch/constants.py

@@ -3,10 +3,6 @@
 import os
 from enum import Enum
 
-from onyx.configs.app_configs import (
-    OPENSEARCH_OVERRIDE_DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES,
-)
-
 
 DEFAULT_MAX_CHUNK_SIZE = 512
 
@@ -41,10 +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
@@ -54,7 +50,7 @@ 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
 
 
 class HybridSearchSubqueryConfiguration(Enum):

backend/onyx/document_index/opensearch/opensearch_document_index.py

@@ -1,12 +1,11 @@
 import json
-from collections.abc import Iterable
+from collections import defaultdict
 from typing import Any
 
 import httpx
 from opensearchpy import NotFoundError
 
 from onyx.access.models import DocumentAccess
-from onyx.configs.app_configs import CHUNKS_PER_BATCH
 from onyx.configs.app_configs import VERIFY_CREATE_OPENSEARCH_INDEX_ON_INIT_MT
 from onyx.configs.chat_configs import NUM_RETURNED_HITS
 from onyx.configs.chat_configs import TITLE_CONTENT_RATIO
@@ -48,6 +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
@@ -118,7 +118,7 @@ def set_cluster_state(client: OpenSearchClient) -> None:
 
 
 def _convert_retrieved_opensearch_chunk_to_inference_chunk_uncleaned(
-    chunk: DocumentChunk,
+    chunk: DocumentChunkWithoutVectors,
     score: float | None,
     highlights: dict[str, list[str]],
 ) -> InferenceChunkUncleaned:
@@ -350,7 +350,7 @@ class OpenSearchOldDocumentIndex(OldDocumentIndex):
 
     def index(
         self,
-        chunks: Iterable[DocMetadataAwareIndexChunk],
+        chunks: list[DocMetadataAwareIndexChunk],
         index_batch_params: IndexBatchParams,
     ) -> set[OldDocumentInsertionRecord]:
         """
@@ -646,8 +646,8 @@ class OpenSearchDocumentIndex(DocumentIndex):
 
     def index(
         self,
-        chunks: Iterable[DocMetadataAwareIndexChunk],
-        indexing_metadata: IndexingMetadata,
+        chunks: list[DocMetadataAwareIndexChunk],
+        indexing_metadata: IndexingMetadata,  # noqa: ARG002
     ) -> list[DocumentInsertionRecord]:
         """Indexes a list of document chunks into the document index.
 
@@ -672,32 +672,29 @@ class OpenSearchDocumentIndex(DocumentIndex):
                 document is newly indexed or had already existed and was just
                 updated.
         """
-        total_chunks = sum(
-            cc.new_chunk_cnt
-            for cc in indexing_metadata.doc_id_to_chunk_cnt_diff.values()
+        # Group chunks by document ID.
+        doc_id_to_chunks: dict[str, list[DocMetadataAwareIndexChunk]] = defaultdict(
+            list
         )
+        for chunk in chunks:
+            doc_id_to_chunks[chunk.source_document.id].append(chunk)
         logger.debug(
-            f"[OpenSearchDocumentIndex] Indexing {total_chunks} chunks from {len(indexing_metadata.doc_id_to_chunk_cnt_diff)} "
+            f"[OpenSearchDocumentIndex] Indexing {len(chunks)} chunks from {len(doc_id_to_chunks)} "
             f"documents for index {self._index_name}."
         )
 
         document_indexing_results: list[DocumentInsertionRecord] = []
-        deleted_doc_ids: set[str] = set()
-        # Buffer chunks per document as they arrive from the iterable.
-        # When the document ID changes flush the buffered chunks.
-        current_doc_id: str | None = None
-        current_chunks: list[DocMetadataAwareIndexChunk] = []
-
-        def _flush_chunks(doc_chunks: list[DocMetadataAwareIndexChunk]) -> None:
+        # Try to index per-document.
+        for _, chunks in doc_id_to_chunks.items():
             # Create a batch of OpenSearch-formatted chunks for bulk insertion.
-            # Since we are doing this in batches, an error occurring midway
-            # can result in a state where chunks are deleted and not all the
-            # new chunks have been indexed.
+            # Do this before deleting existing chunks to reduce the amount of
+            # time the document index has no content for a given document, and
+            # to reduce the chance of entering a state where we delete chunks,
+            # then some error happens, and never successfully index new chunks.
             chunk_batch: list[DocumentChunk] = [
-                _convert_onyx_chunk_to_opensearch_document(chunk)
-                for chunk in doc_chunks
+                _convert_onyx_chunk_to_opensearch_document(chunk) for chunk in chunks
             ]
-            onyx_document: Document = doc_chunks[0].source_document
+            onyx_document: Document = chunks[0].source_document
             # First delete the doc's chunks from the index. This is so that
             # there are no dangling chunks in the index, in the event that the
             # new document's content contains fewer chunks than the previous
@@ -706,43 +703,22 @@ class OpenSearchDocumentIndex(DocumentIndex):
             # if the chunk count has actually decreased. This assumes that
             # overlapping chunks are perfectly overwritten. If we can't
             # guarantee that then we need the code as-is.
-            if onyx_document.id not in deleted_doc_ids:
-                num_chunks_deleted = self.delete(
-                    onyx_document.id, onyx_document.chunk_count
-                )
-                deleted_doc_ids.add(onyx_document.id)
-                # If we see that chunks were deleted we assume the doc already
-                # existed. We record the result before bulk_index_documents
-                # runs. If indexing raises, this entire result list is discarded
-                # by the caller's retry logic, so early recording is safe.
-                document_indexing_results.append(
-                    DocumentInsertionRecord(
-                        document_id=onyx_document.id,
-                        already_existed=num_chunks_deleted > 0,
-                    )
-                )
+            num_chunks_deleted = self.delete(
+                onyx_document.id, onyx_document.chunk_count
+            )
+            # If we see that chunks were deleted we assume the doc already
+            # existed.
+            document_insertion_record = DocumentInsertionRecord(
+                document_id=onyx_document.id,
+                already_existed=num_chunks_deleted > 0,
+            )
             # Now index. This will raise if a chunk of the same ID exists, which
             # we do not expect because we should have deleted all chunks.
             self._client.bulk_index_documents(
                 documents=chunk_batch,
                 tenant_state=self._tenant_state,
             )
-
-        for chunk in chunks:
-            doc_id = chunk.source_document.id
-            if doc_id != current_doc_id:
-                if current_chunks:
-                    _flush_chunks(current_chunks)
-                current_doc_id = doc_id
-                current_chunks = [chunk]
-            elif len(current_chunks) >= CHUNKS_PER_BATCH:
-                _flush_chunks(current_chunks)
-                current_chunks = [chunk]
-            else:
-                current_chunks.append(chunk)
-
-        if current_chunks:
-            _flush_chunks(current_chunks)
+            document_indexing_results.append(document_insertion_record)
 
         return document_indexing_results
 
@@ -905,7 +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,
@@ -969,7 +945,7 @@ class OpenSearchDocumentIndex(DocumentIndex):
             include_hidden=False,
         )
         normalization_pipeline_name, _ = get_normalization_pipeline_name_and_config()
-        search_hits: list[SearchHit[DocumentChunk]] = self._client.search(
+        search_hits: list[SearchHit[DocumentChunkWithoutVectors]] = self._client.search(
             body=query_body,
             search_pipeline_id=normalization_pipeline_name,
         )
@@ -1001,7 +977,7 @@ 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

@@ -7,6 +7,7 @@ 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
@@ -15,7 +16,7 @@ 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 (
     DEFAULT_OPENSEARCH_MAX_RESULT_WINDOW,
@@ -235,9 +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
 
@@ -332,7 +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
@@ -356,9 +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] = {
@@ -385,10 +391,19 @@ 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]
+            },
         }
 
+        if not OPENSEARCH_MATCH_HIGHLIGHTS_DISABLED:
+            final_hybrid_search_body["highlight"] = (
+                DocumentQuery._get_match_highlights_configuration()
+            )
+
         # Explain is for scoring breakdowns.
         if OPENSEARCH_EXPLAIN_ENABLED:
             final_hybrid_search_body["explain"] = True
@@ -446,6 +461,11 @@ 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
@@ -460,7 +480,7 @@ 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.
 
@@ -546,7 +566,7 @@ class DocumentQuery:
     @staticmethod
     def _get_title_vector_similarity_search_query(
         query_vector: list[float],
-        vector_candidates: int = DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES,
+        vector_candidates: int = DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES,
     ) -> dict[str, Any]:
         return {
             "knn": {
@@ -560,7 +580,7 @@ class DocumentQuery:
     @staticmethod
     def _get_content_vector_similarity_search_query(
         query_vector: list[float],
-        vector_candidates: int = DEFAULT_NUM_HYBRID_SEARCH_CANDIDATES,
+        vector_candidates: int = DEFAULT_NUM_HYBRID_SUBQUERY_CANDIDATES,
     ) -> dict[str, Any]:
         return {
             "knn": {

backend/onyx/document_index/vespa/index.py

@@ -6,7 +6,6 @@ import re
 import time
 import urllib
 import zipfile
-from collections.abc import Iterable
 from dataclasses import dataclass
 from datetime import datetime
 from datetime import timedelta
@@ -462,7 +461,7 @@ class VespaIndex(DocumentIndex):
 
     def index(
         self,
-        chunks: Iterable[DocMetadataAwareIndexChunk],
+        chunks: list[DocMetadataAwareIndexChunk],
         index_batch_params: IndexBatchParams,
     ) -> set[OldDocumentInsertionRecord]:
         """

backend/onyx/document_index/vespa/vespa_document_index.py

@@ -1,8 +1,6 @@
 import concurrent.futures
 import logging
 import random
-from collections.abc import Generator
-from collections.abc import Iterable
 from typing import Any
 from uuid import UUID
 
@@ -10,7 +8,6 @@ import httpx
 from pydantic import BaseModel
 from retry import retry
 
-from onyx.configs.app_configs import CHUNKS_PER_BATCH
 from onyx.configs.app_configs import RECENCY_BIAS_MULTIPLIER
 from onyx.configs.app_configs import RERANK_COUNT
 from onyx.configs.chat_configs import DOC_TIME_DECAY
@@ -321,7 +318,7 @@ class VespaDocumentIndex(DocumentIndex):
 
     def index(
         self,
-        chunks: Iterable[DocMetadataAwareIndexChunk],
+        chunks: list[DocMetadataAwareIndexChunk],
         indexing_metadata: IndexingMetadata,
     ) -> list[DocumentInsertionRecord]:
         doc_id_to_chunk_cnt_diff = indexing_metadata.doc_id_to_chunk_cnt_diff
@@ -341,31 +338,22 @@ class VespaDocumentIndex(DocumentIndex):
 
         # Vespa has restrictions on valid characters, yet document IDs come from
         # external w.r.t. this class. We need to sanitize them.
-        #
-        # Instead of materializing all cleaned chunks upfront, we stream them
-        # through a generator that cleans IDs and builds the original-ID mapping
-        # incrementally as chunks flow into Vespa.
-        def _clean_and_track(
-            chunks_iter: Iterable[DocMetadataAwareIndexChunk],
-            id_map: dict[str, str],
-            seen_ids: set[str],
-        ) -> Generator[DocMetadataAwareIndexChunk, None, None]:
-            """Cleans chunk IDs and builds the original-ID mapping
-            incrementally as chunks flow through, avoiding a separate
-            materialization pass."""
-            for chunk in chunks_iter:
-                original_id = chunk.source_document.id
-                cleaned = clean_chunk_id_copy(chunk)
-                cleaned_id = cleaned.source_document.id
-                # Needed so the final DocumentInsertionRecord returned can have
-                # the original document ID. cleaned_chunks might not contain IDs
-                # exactly as callers supplied them.
-                id_map[cleaned_id] = original_id
-                seen_ids.add(cleaned_id)
-                yield cleaned
+        cleaned_chunks: list[DocMetadataAwareIndexChunk] = [
+            clean_chunk_id_copy(chunk) for chunk in chunks
+        ]
+        assert len(cleaned_chunks) == len(
+            chunks
+        ), "Bug: Cleaned chunks and input chunks have different lengths."
 
-        new_document_id_to_original_document_id: dict[str, str] = {}
-        all_cleaned_doc_ids: set[str] = set()
+        # Needed so the final DocumentInsertionRecord returned can have the
+        # original document ID. cleaned_chunks might not contain IDs exactly as
+        # callers supplied them.
+        new_document_id_to_original_document_id: dict[str, str] = dict()
+        for i, cleaned_chunk in enumerate(cleaned_chunks):
+            old_chunk = chunks[i]
+            new_document_id_to_original_document_id[
+                cleaned_chunk.source_document.id
+            ] = old_chunk.source_document.id
 
         existing_docs: set[str] = set()
 
@@ -421,16 +409,8 @@ class VespaDocumentIndex(DocumentIndex):
                     executor=executor,
                 )
 
-            # Insert new Vespa documents, streaming through the cleaning
-            # pipeline so chunks are never fully materialized.
-            cleaned_chunks = _clean_and_track(
-                chunks,
-                new_document_id_to_original_document_id,
-                all_cleaned_doc_ids,
-            )
-            for chunk_batch in batch_generator(
-                cleaned_chunks, min(BATCH_SIZE, CHUNKS_PER_BATCH)
-            ):
+            # Insert new Vespa documents.
+            for chunk_batch in batch_generator(cleaned_chunks, BATCH_SIZE):
                 batch_index_vespa_chunks(
                     chunks=chunk_batch,
                     index_name=self._index_name,
@@ -439,6 +419,10 @@ class VespaDocumentIndex(DocumentIndex):
                     executor=executor,
                 )
 
+        all_cleaned_doc_ids: set[str] = {
+            chunk.source_document.id for chunk in cleaned_chunks
+        }
+
         return [
             DocumentInsertionRecord(
                 document_id=new_document_id_to_original_document_id[cleaned_doc_id],

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/models.py

@@ -0,0 +1,127 @@
+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  # if None in model_fields_set, reset to spec default
+    )
+    timeout_seconds: float | None = Field(
+        default=None, gt=0
+    )  # if None in model_fields_set, reset to spec default
+
+    @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.")
+        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
+    creator_email: str | None
+    created_at: datetime
+    updated_at: datetime
+
+
+class HookValidateResponse(BaseModel):
+    success: bool
+    error_message: str | None = None
+
+
+# ---------------------------------------------------------------------------
+# Health models
+# ---------------------------------------------------------------------------
+
+
+class HookHealthStatus(str, Enum):
+    healthy = "healthy"  # green — reachable, no failures in last 1h
+    degraded = "degraded"  # yellow — reachable, failures in last 1h
+    unreachable = "unreachable"  # red — is_reachable=false or null
+
+
+class HookFailureRecord(BaseModel):
+    error_message: str | None = None
+    status_code: int | None = None
+    duration_ms: int | None = None
+    created_at: datetime
+
+
+class HookHealthResponse(BaseModel):
+    status: HookHealthStatus
+    recent_failures: list[HookFailureRecord] = Field(
+        default_factory=list,
+        description="Last 10 failures, newest first",
+        max_length=10,
+    )

backend/onyx/hooks/points/base.py

@@ -0,0 +1,59 @@
+from abc import ABC
+from abc import abstractmethod
+from typing import Any
+
+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",
+)
+
+
+class HookPointSpec(ABC):
+    """Static metadata and contract for a pipeline hook point.
+
+    This is NOT a regular class meant for direct instantiation by callers.
+    Each concrete subclass represents exactly one hook point and is instantiated
+    once at startup, registered in onyx.hooks.registry._REGISTRY. No caller
+    should ever create instances directly — use get_hook_point_spec() or
+    get_all_specs() from the registry instead.
+
+    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.
+    """
+
+    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
+
+    def __init_subclass__(cls, **kwargs: object) -> None:
+        super().__init_subclass__(**kwargs)
+        # Skip intermediate abstract subclasses — they may still be partially defined.
+        if getattr(cls, "__abstractmethods__", None):
+            return
+        missing = [attr for attr in _REQUIRED_ATTRS if not hasattr(cls, attr)]
+        if missing:
+            raise TypeError(f"{cls.__name__} must define class attributes: {missing}")
+
+    @property
+    @abstractmethod
+    def input_schema(self) -> dict[str, Any]:
+        """JSON schema describing the request payload sent to the customer's endpoint."""
+
+    @property
+    @abstractmethod
+    def output_schema(self) -> dict[str, Any]:
+        """JSON schema describing the expected response from the customer's endpoint."""

backend/onyx/hooks/points/document_ingestion.py

@@ -0,0 +1,29 @@
+from typing import Any
+
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.hooks.points.base import HookPointSpec
+
+
+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
+
+    @property
+    def input_schema(self) -> dict[str, Any]:
+        # TODO(@Bo-Onyx): define input schema
+        return {"type": "object", "properties": {}}
+
+    @property
+    def output_schema(self) -> dict[str, Any]:
+        # TODO(@Bo-Onyx): define output schema
+        return {"type": "object", "properties": {}}

backend/onyx/hooks/points/query_processing.py

@@ -0,0 +1,83 @@
+from typing import Any
+
+from onyx.db.enums import HookFailStrategy
+from onyx.db.enums import HookPoint
+from onyx.hooks.points.base import HookPointSpec
+
+
+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
+
+    @property
+    def input_schema(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "query": {
+                    "type": "string",
+                    "description": "The raw query string exactly as the user typed it.",
+                },
+                "user_email": {
+                    "type": ["string", "null"],
+                    "description": "Email of the user submitting the query, or null if unauthenticated.",
+                },
+                "chat_session_id": {
+                    "type": "string",
+                    "description": "UUID of the chat session. Always present — the session is guaranteed to exist by the time this hook fires.",
+                },
+            },
+            "required": ["query", "user_email", "chat_session_id"],
+            "additionalProperties": False,
+        }
+
+    @property
+    def output_schema(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "query": {
+                    "type": ["string", "null"],
+                    "description": (
+                        "The (optionally modified) query to use. "
+                        "Set to null to reject the query."
+                    ),
+                },
+                "rejection_message": {
+                    "type": ["string", "null"],
+                    "description": (
+                        "Message shown to the user when query is null. "
+                        "Falls back to a generic message if not provided."
+                    ),
+                },
+            },
+            "required": ["query"],
+        }

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/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/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
@@ -308,6 +309,7 @@ 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:

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/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/requirements/default.txt

@@ -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/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

@@ -29,6 +29,7 @@ from onyx.document_index.opensearch.opensearch_document_index import (
 )
 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
@@ -96,6 +97,23 @@ def _patch_hybrid_search_normalization_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,
@@ -226,7 +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.
@@ -242,7 +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.
@@ -271,7 +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)
 
@@ -285,7 +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.
@@ -305,7 +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.
@@ -340,7 +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.
@@ -383,7 +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.
@@ -418,7 +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)
 
@@ -461,7 +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(
@@ -489,7 +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 = [
@@ -520,7 +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(
@@ -548,7 +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(
@@ -583,7 +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.
@@ -602,7 +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(
@@ -638,7 +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.
@@ -659,7 +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.
@@ -735,7 +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.
@@ -784,7 +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.
@@ -804,11 +822,12 @@ class TestOpenSearchClient:
         """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 = {
@@ -881,8 +900,12 @@ class TestOpenSearchClient:
                 )
                 # Make sure the chunk contents are preserved.
                 for i, chunk in enumerate(results):
-                    assert (
-                        chunk.document_chunk == docs[chunk.document_chunk.document_id]
+                    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).
@@ -906,7 +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.
 
@@ -942,12 +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.
@@ -1038,7 +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
@@ -1046,7 +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, [])
 
@@ -1062,11 +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.
@@ -1193,7 +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
@@ -1286,7 +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.
@@ -1313,7 +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.
@@ -1381,7 +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.
@@ -1458,7 +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.
@@ -1550,7 +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.
@@ -1599,4 +1634,9 @@ 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
+                }
+            )

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/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/document_index/opensearch/test_opensearch_batch_flush.py

@@ -1,208 +0,0 @@
-from unittest.mock import MagicMock
-from unittest.mock import patch
-
-from onyx.access.models import DocumentAccess
-from onyx.configs.constants import DocumentSource
-from onyx.connectors.models import Document
-from onyx.connectors.models import TextSection
-from onyx.document_index.interfaces_new import IndexingMetadata
-from onyx.document_index.interfaces_new import TenantState
-from onyx.document_index.opensearch.opensearch_document_index import (
-    OpenSearchDocumentIndex,
-)
-from onyx.indexing.models import DocMetadataAwareIndexChunk
-
-
-def _make_chunk(
-    doc_id: str,
-    chunk_id: int,
-) -> DocMetadataAwareIndexChunk:
-    """Creates a minimal DocMetadataAwareIndexChunk for testing."""
-    doc = Document(
-        id=doc_id,
-        sections=[TextSection(text="test", link="http://test.com")],
-        source=DocumentSource.FILE,
-        semantic_identifier="test_doc",
-        metadata={},
-    )
-    access = DocumentAccess.build(
-        user_emails=[],
-        user_groups=[],
-        external_user_emails=[],
-        external_user_group_ids=[],
-        is_public=True,
-    )
-    return DocMetadataAwareIndexChunk(
-        chunk_id=chunk_id,
-        blurb="test",
-        content="test content",
-        source_links={0: "http://test.com"},
-        image_file_id=None,
-        section_continuation=False,
-        source_document=doc,
-        title_prefix="",
-        metadata_suffix_semantic="",
-        metadata_suffix_keyword="",
-        mini_chunk_texts=None,
-        large_chunk_id=None,
-        doc_summary="",
-        chunk_context="",
-        contextual_rag_reserved_tokens=0,
-        embeddings={"full_embedding": [0.1] * 10, "mini_chunk_embeddings": []},
-        title_embedding=[0.1] * 10,
-        tenant_id="test_tenant",
-        access=access,
-        document_sets=set(),
-        user_project=[],
-        personas=[],
-        boost=0,
-        aggregated_chunk_boost_factor=1.0,
-        ancestor_hierarchy_node_ids=[],
-    )
-
-
-def _make_index() -> OpenSearchDocumentIndex:
-    """Creates an OpenSearchDocumentIndex with a mocked client."""
-    mock_client = MagicMock()
-    mock_client.bulk_index_documents = MagicMock()
-
-    tenant_state = TenantState(tenant_id="test_tenant", multitenant=False)
-
-    index = OpenSearchDocumentIndex.__new__(OpenSearchDocumentIndex)
-    index._index_name = "test_index"
-    index._client = mock_client
-    index._tenant_state = tenant_state
-
-    return index
-
-
-def _make_metadata(doc_id: str, chunk_count: int) -> IndexingMetadata:
-    return IndexingMetadata(
-        doc_id_to_chunk_cnt_diff={
-            doc_id: IndexingMetadata.ChunkCounts(
-                old_chunk_cnt=0,
-                new_chunk_cnt=chunk_count,
-            ),
-        },
-    )
-
-
-@patch("onyx.document_index.opensearch.opensearch_document_index.CHUNKS_PER_BATCH", 100)
-def test_single_doc_under_batch_limit_flushes_once() -> None:
-    """A document with fewer chunks than CHUNKS_PER_BATCH should flush once."""
-    index = _make_index()
-    doc_id = "doc_1"
-    num_chunks = 50
-    chunks = [_make_chunk(doc_id, i) for i in range(num_chunks)]
-    metadata = _make_metadata(doc_id, num_chunks)
-
-    with patch.object(index, "delete", return_value=0):
-        index.index(chunks, metadata)
-
-    assert index._client.bulk_index_documents.call_count == 1
-    batch_arg = index._client.bulk_index_documents.call_args_list[0]
-    assert len(batch_arg.kwargs["documents"]) == num_chunks
-
-
-@patch("onyx.document_index.opensearch.opensearch_document_index.CHUNKS_PER_BATCH", 100)
-def test_single_doc_over_batch_limit_flushes_multiple_times() -> None:
-    """A document with more chunks than CHUNKS_PER_BATCH should flush multiple times."""
-    index = _make_index()
-    doc_id = "doc_1"
-    num_chunks = 250
-    chunks = [_make_chunk(doc_id, i) for i in range(num_chunks)]
-    metadata = _make_metadata(doc_id, num_chunks)
-
-    with patch.object(index, "delete", return_value=0):
-        index.index(chunks, metadata)
-
-    # 250 chunks / 100 per batch = 3 flushes (100 + 100 + 50)
-    assert index._client.bulk_index_documents.call_count == 3
-    batch_sizes = [
-        len(call.kwargs["documents"])
-        for call in index._client.bulk_index_documents.call_args_list
-    ]
-    assert batch_sizes == [100, 100, 50]
-
-
-@patch("onyx.document_index.opensearch.opensearch_document_index.CHUNKS_PER_BATCH", 100)
-def test_single_doc_exactly_at_batch_limit() -> None:
-    """A document with exactly CHUNKS_PER_BATCH chunks should flush once
-    (the flush happens on the next chunk, not at the boundary)."""
-    index = _make_index()
-    doc_id = "doc_1"
-    num_chunks = 100
-    chunks = [_make_chunk(doc_id, i) for i in range(num_chunks)]
-    metadata = _make_metadata(doc_id, num_chunks)
-
-    with patch.object(index, "delete", return_value=0):
-        index.index(chunks, metadata)
-
-    # 100 chunks hit the >= check on chunk 101 which doesn't exist,
-    # so final flush handles all 100
-    # Actually: the elif fires when len(current_chunks) >= 100, which happens
-    # when current_chunks has 100 items and the 101st chunk arrives.
-    # With exactly 100 chunks, the 100th chunk makes len == 99, then appended -> 100.
-    # No 101st chunk arrives, so the final flush handles all 100.
-    assert index._client.bulk_index_documents.call_count == 1
-
-
-@patch("onyx.document_index.opensearch.opensearch_document_index.CHUNKS_PER_BATCH", 100)
-def test_single_doc_one_over_batch_limit() -> None:
-    """101 chunks for one doc: first 100 flushed when the 101st arrives, then
-    the 101st is flushed at the end."""
-    index = _make_index()
-    doc_id = "doc_1"
-    num_chunks = 101
-    chunks = [_make_chunk(doc_id, i) for i in range(num_chunks)]
-    metadata = _make_metadata(doc_id, num_chunks)
-
-    with patch.object(index, "delete", return_value=0):
-        index.index(chunks, metadata)
-
-    assert index._client.bulk_index_documents.call_count == 2
-    batch_sizes = [
-        len(call.kwargs["documents"])
-        for call in index._client.bulk_index_documents.call_args_list
-    ]
-    assert batch_sizes == [100, 1]
-
-
-@patch("onyx.document_index.opensearch.opensearch_document_index.CHUNKS_PER_BATCH", 100)
-def test_multiple_docs_each_under_limit_flush_per_doc() -> None:
-    """Multiple documents each under the batch limit should flush once per document."""
-    index = _make_index()
-    chunks = []
-    for doc_idx in range(3):
-        doc_id = f"doc_{doc_idx}"
-        for chunk_idx in range(50):
-            chunks.append(_make_chunk(doc_id, chunk_idx))
-
-    metadata = IndexingMetadata(
-        doc_id_to_chunk_cnt_diff={
-            f"doc_{i}": IndexingMetadata.ChunkCounts(old_chunk_cnt=0, new_chunk_cnt=50)
-            for i in range(3)
-        },
-    )
-
-    with patch.object(index, "delete", return_value=0):
-        index.index(chunks, metadata)
-
-    # 3 documents = 3 flushes (one per doc boundary + final)
-    assert index._client.bulk_index_documents.call_count == 3
-
-
-@patch("onyx.document_index.opensearch.opensearch_document_index.CHUNKS_PER_BATCH", 100)
-def test_delete_called_once_per_document() -> None:
-    """Even with multiple flushes for a single document, delete should only be
-    called once per document."""
-    index = _make_index()
-    doc_id = "doc_1"
-    num_chunks = 250
-    chunks = [_make_chunk(doc_id, i) for i in range(num_chunks)]
-    metadata = _make_metadata(doc_id, num_chunks)
-
-    with patch.object(index, "delete", return_value=0) as mock_delete:
-        index.index(chunks, metadata)
-
-    mock_delete.assert_called_once_with(doc_id, None)

backend/tests/unit/onyx/document_index/opensearch/test_opensearch_index.py

@@ -1,272 +0,0 @@
-"""Unit tests for OpenSearchDocumentIndex.index().
-
-These tests mock the OpenSearch client and verify the buffered
-flush-by-document logic, DocumentInsertionRecord construction, and
-delete-before-insert semantics.
-"""
-
-from collections.abc import Iterator
-from unittest.mock import MagicMock
-from unittest.mock import patch
-
-from onyx.access.models import DocumentAccess
-from onyx.connectors.models import Document
-from onyx.connectors.models import DocumentSource
-from onyx.connectors.models import TextSection
-from onyx.document_index.interfaces_new import IndexingMetadata
-from onyx.document_index.interfaces_new import TenantState
-from onyx.document_index.opensearch.opensearch_document_index import (
-    OpenSearchDocumentIndex,
-)
-from onyx.indexing.models import ChunkEmbedding
-from onyx.indexing.models import DocMetadataAwareIndexChunk
-from onyx.indexing.models import IndexChunk
-
-
-def _make_chunk(
-    doc_id: str,
-    chunk_id: int = 0,
-    content: str = "test content",
-    chunk_count: int | None = None,
-) -> DocMetadataAwareIndexChunk:
-    doc = Document(
-        id=doc_id,
-        semantic_identifier="test_doc",
-        sections=[TextSection(text=content, link=None)],
-        source=DocumentSource.NOT_APPLICABLE,
-        metadata={},
-        chunk_count=chunk_count,
-    )
-    index_chunk = IndexChunk(
-        chunk_id=chunk_id,
-        blurb=content[:50],
-        content=content,
-        source_links=None,
-        image_file_id=None,
-        section_continuation=False,
-        source_document=doc,
-        title_prefix="",
-        metadata_suffix_semantic="",
-        metadata_suffix_keyword="",
-        contextual_rag_reserved_tokens=0,
-        doc_summary="",
-        chunk_context="",
-        mini_chunk_texts=None,
-        large_chunk_id=None,
-        embeddings=ChunkEmbedding(
-            full_embedding=[0.1] * 10,
-            mini_chunk_embeddings=[],
-        ),
-        title_embedding=[0.1] * 10,
-    )
-    access = DocumentAccess.build(
-        user_emails=[],
-        user_groups=[],
-        external_user_emails=[],
-        external_user_group_ids=[],
-        is_public=True,
-    )
-    return DocMetadataAwareIndexChunk.from_index_chunk(
-        index_chunk=index_chunk,
-        access=access,
-        document_sets=set(),
-        user_project=[],
-        personas=[],
-        boost=0,
-        aggregated_chunk_boost_factor=1.0,
-        tenant_id="test_tenant",
-    )
-
-
-def _make_indexing_metadata(
-    doc_ids: list[str],
-    old_counts: list[int],
-    new_counts: list[int],
-) -> IndexingMetadata:
-    return IndexingMetadata(
-        doc_id_to_chunk_cnt_diff={
-            doc_id: IndexingMetadata.ChunkCounts(
-                old_chunk_cnt=old,
-                new_chunk_cnt=new,
-            )
-            for doc_id, old, new in zip(doc_ids, old_counts, new_counts)
-        }
-    )
-
-
-def _make_os_index(mock_client: MagicMock) -> OpenSearchDocumentIndex:
-    """Create an OpenSearchDocumentIndex with a mocked client."""
-    with patch.object(
-        OpenSearchDocumentIndex,
-        "__init__",
-        lambda _self, *_a, **_kw: None,
-    ):
-        idx = OpenSearchDocumentIndex.__new__(OpenSearchDocumentIndex)
-
-    idx._index_name = "test_index"
-    idx._tenant_state = TenantState(tenant_id="test_tenant", multitenant=False)
-    idx._client = mock_client
-    return idx
-
-
-def test_index_single_new_doc() -> None:
-    """Indexing a single new document returns one record with already_existed=False."""
-    mock_client = MagicMock()
-    mock_client.bulk_index_documents.return_value = None
-
-    idx = _make_os_index(mock_client)
-
-    # Patch delete to return 0 (no existing chunks)
-    with patch.object(idx, "delete", return_value=0) as mock_delete:
-        chunk = _make_chunk("doc1")
-        metadata = _make_indexing_metadata(["doc1"], old_counts=[0], new_counts=[1])
-
-        results = idx.index(chunks=[chunk], indexing_metadata=metadata)
-
-    assert len(results) == 1
-    assert results[0].document_id == "doc1"
-    assert results[0].already_existed is False
-    mock_delete.assert_called_once()
-    mock_client.bulk_index_documents.assert_called_once()
-
-
-def test_index_existing_doc_already_existed_true() -> None:
-    """Re-indexing a doc with previous chunks returns already_existed=True."""
-    mock_client = MagicMock()
-    mock_client.bulk_index_documents.return_value = None
-
-    idx = _make_os_index(mock_client)
-
-    with patch.object(idx, "delete", return_value=5):
-        chunk = _make_chunk("doc1")
-        metadata = _make_indexing_metadata(["doc1"], old_counts=[5], new_counts=[1])
-
-        results = idx.index(chunks=[chunk], indexing_metadata=metadata)
-
-    assert len(results) == 1
-    assert results[0].already_existed is True
-
-
-def test_index_multiple_docs_flushed_separately() -> None:
-    """Chunks from different documents are flushed in separate bulk calls."""
-    mock_client = MagicMock()
-    mock_client.bulk_index_documents.return_value = None
-
-    idx = _make_os_index(mock_client)
-
-    with patch.object(idx, "delete", return_value=0):
-        chunks = [
-            _make_chunk("doc1", chunk_id=0),
-            _make_chunk("doc1", chunk_id=1),
-            _make_chunk("doc2", chunk_id=0),
-        ]
-        metadata = _make_indexing_metadata(
-            ["doc1", "doc2"], old_counts=[0, 0], new_counts=[2, 1]
-        )
-
-        results = idx.index(chunks=chunks, indexing_metadata=metadata)
-
-    result_map = {r.document_id: r.already_existed for r in results}
-    assert len(result_map) == 2
-    assert result_map["doc1"] is False
-    assert result_map["doc2"] is False
-    # Two separate flushes: one for doc1 (2 chunks), one for doc2 (1 chunk)
-    assert mock_client.bulk_index_documents.call_count == 2
-
-
-def test_index_deletes_before_inserting() -> None:
-    """For each document, delete is called before bulk_index_documents."""
-    mock_client = MagicMock()
-    mock_client.bulk_index_documents.return_value = None
-
-    call_order: list[str] = []
-
-    idx = _make_os_index(mock_client)
-
-    def track_delete(*_args: object, **_kwargs: object) -> int:
-        call_order.append("delete")
-        return 3
-
-    def track_bulk(*_args: object, **_kwargs: object) -> None:
-        call_order.append("bulk_index")
-
-    mock_client.bulk_index_documents.side_effect = track_bulk
-
-    with patch.object(idx, "delete", side_effect=track_delete):
-        chunk = _make_chunk("doc1")
-        metadata = _make_indexing_metadata(["doc1"], old_counts=[3], new_counts=[1])
-
-        idx.index(chunks=[chunk], indexing_metadata=metadata)
-
-    assert call_order == ["delete", "bulk_index"]
-
-
-def test_index_delete_called_once_per_doc() -> None:
-    """Delete is called only once per document, even with multiple chunks."""
-    mock_client = MagicMock()
-    mock_client.bulk_index_documents.return_value = None
-
-    idx = _make_os_index(mock_client)
-
-    with patch.object(idx, "delete", return_value=0) as mock_delete:
-        # 3 chunks, all same doc — should only delete once
-        chunks = [_make_chunk("doc1", chunk_id=i) for i in range(3)]
-        metadata = _make_indexing_metadata(["doc1"], old_counts=[0], new_counts=[3])
-
-        idx.index(chunks=chunks, indexing_metadata=metadata)
-
-    mock_delete.assert_called_once()
-
-
-def test_index_flushes_on_doc_boundary() -> None:
-    """When doc ID changes in the stream, the previous doc's chunks are flushed."""
-    mock_client = MagicMock()
-    mock_client.bulk_index_documents.return_value = None
-
-    idx = _make_os_index(mock_client)
-
-    bulk_call_chunk_counts: list[int] = []
-
-    def track_bulk(documents: list[object], **_kwargs: object) -> None:
-        bulk_call_chunk_counts.append(len(documents))
-
-    mock_client.bulk_index_documents.side_effect = track_bulk
-
-    with patch.object(idx, "delete", return_value=0):
-        chunks = [
-            _make_chunk("doc1", chunk_id=0),
-            _make_chunk("doc1", chunk_id=1),
-            _make_chunk("doc1", chunk_id=2),
-            _make_chunk("doc2", chunk_id=0),
-            _make_chunk("doc2", chunk_id=1),
-        ]
-        metadata = _make_indexing_metadata(
-            ["doc1", "doc2"], old_counts=[0, 0], new_counts=[3, 2]
-        )
-
-        idx.index(chunks=chunks, indexing_metadata=metadata)
-
-    # First flush: 3 chunks for doc1, second flush: 2 chunks for doc2
-    assert bulk_call_chunk_counts == [3, 2]
-
-
-def test_index_with_generator_input() -> None:
-    """The index method works with a generator (iterable) input, not just lists."""
-    mock_client = MagicMock()
-    mock_client.bulk_index_documents.return_value = None
-
-    idx = _make_os_index(mock_client)
-
-    consumed: list[int] = []
-
-    def chunk_gen() -> Iterator[DocMetadataAwareIndexChunk]:
-        for i in range(3):
-            consumed.append(i)
-            yield _make_chunk("doc1", chunk_id=i)
-
-    with patch.object(idx, "delete", return_value=0):
-        metadata = _make_indexing_metadata(["doc1"], old_counts=[0], new_counts=[3])
-        results = idx.index(chunks=chunk_gen(), indexing_metadata=metadata)
-
-    assert consumed == [0, 1, 2]
-    assert len(results) == 1

backend/tests/unit/onyx/document_index/vespa/test_vespa_index.py

@@ -1,417 +0,0 @@
-"""Unit tests for VespaDocumentIndex.index().
-
-These tests mock all external I/O (HTTP calls, thread pools) and verify
-the streaming logic, ID cleaning/mapping, and DocumentInsertionRecord
-construction.
-"""
-
-from collections.abc import Iterator
-from unittest.mock import MagicMock
-from unittest.mock import patch
-from uuid import uuid4
-
-from onyx.access.models import DocumentAccess
-from onyx.connectors.models import Document
-from onyx.connectors.models import DocumentSource
-from onyx.connectors.models import TextSection
-from onyx.document_index.interfaces import EnrichedDocumentIndexingInfo
-from onyx.document_index.interfaces_new import IndexingMetadata
-from onyx.document_index.interfaces_new import TenantState
-from onyx.document_index.vespa.vespa_document_index import VespaDocumentIndex
-from onyx.indexing.models import ChunkEmbedding
-from onyx.indexing.models import DocMetadataAwareIndexChunk
-from onyx.indexing.models import IndexChunk
-
-
-def _make_chunk(
-    doc_id: str,
-    chunk_id: int = 0,
-    content: str = "test content",
-) -> DocMetadataAwareIndexChunk:
-    doc = Document(
-        id=doc_id,
-        semantic_identifier="test_doc",
-        sections=[TextSection(text=content, link=None)],
-        source=DocumentSource.NOT_APPLICABLE,
-        metadata={},
-    )
-    index_chunk = IndexChunk(
-        chunk_id=chunk_id,
-        blurb=content[:50],
-        content=content,
-        source_links=None,
-        image_file_id=None,
-        section_continuation=False,
-        source_document=doc,
-        title_prefix="",
-        metadata_suffix_semantic="",
-        metadata_suffix_keyword="",
-        contextual_rag_reserved_tokens=0,
-        doc_summary="",
-        chunk_context="",
-        mini_chunk_texts=None,
-        large_chunk_id=None,
-        embeddings=ChunkEmbedding(
-            full_embedding=[0.1] * 10,
-            mini_chunk_embeddings=[],
-        ),
-        title_embedding=None,
-    )
-    access = DocumentAccess.build(
-        user_emails=[],
-        user_groups=[],
-        external_user_emails=[],
-        external_user_group_ids=[],
-        is_public=True,
-    )
-    return DocMetadataAwareIndexChunk.from_index_chunk(
-        index_chunk=index_chunk,
-        access=access,
-        document_sets=set(),
-        user_project=[],
-        personas=[],
-        boost=0,
-        aggregated_chunk_boost_factor=1.0,
-        tenant_id="test_tenant",
-    )
-
-
-def _make_indexing_metadata(
-    doc_ids: list[str],
-    old_counts: list[int],
-    new_counts: list[int],
-) -> IndexingMetadata:
-    return IndexingMetadata(
-        doc_id_to_chunk_cnt_diff={
-            doc_id: IndexingMetadata.ChunkCounts(
-                old_chunk_cnt=old,
-                new_chunk_cnt=new,
-            )
-            for doc_id, old, new in zip(doc_ids, old_counts, new_counts)
-        }
-    )
-
-
-def _stub_enrich(
-    doc_id: str,
-    old_chunk_cnt: int,
-) -> EnrichedDocumentIndexingInfo:
-    """Build an EnrichedDocumentIndexingInfo that says 'no chunks to delete'
-    when old_chunk_cnt == 0, or 'has existing chunks' otherwise."""
-    return EnrichedDocumentIndexingInfo(
-        doc_id=doc_id,
-        chunk_start_index=0,
-        old_version=False,
-        chunk_end_index=old_chunk_cnt,
-    )
-
-
-@patch("onyx.document_index.vespa.vespa_document_index.batch_index_vespa_chunks")
-@patch("onyx.document_index.vespa.vespa_document_index.delete_vespa_chunks")
-@patch(
-    "onyx.document_index.vespa.vespa_document_index.get_document_chunk_ids",
-    return_value=[],
-)
-@patch("onyx.document_index.vespa.vespa_document_index._enrich_basic_chunk_info")
-def test_index_single_new_doc(
-    mock_enrich: MagicMock,
-    mock_get_chunk_ids: MagicMock,  # noqa: ARG001
-    mock_delete: MagicMock,  # noqa: ARG001
-    mock_batch_index: MagicMock,
-) -> None:
-    """Indexing a single new document returns one record with already_existed=False."""
-    mock_enrich.return_value = _stub_enrich("doc1", old_chunk_cnt=0)
-
-    index = VespaDocumentIndex(
-        index_name="test_index",
-        tenant_state=TenantState(tenant_id="test_tenant", multitenant=False),
-        large_chunks_enabled=False,
-        httpx_client=MagicMock(),
-    )
-
-    chunk = _make_chunk("doc1")
-    metadata = _make_indexing_metadata(["doc1"], old_counts=[0], new_counts=[1])
-
-    results = index.index(chunks=[chunk], indexing_metadata=metadata)
-
-    assert len(results) == 1
-    assert results[0].document_id == "doc1"
-    assert results[0].already_existed is False
-
-    # batch_index_vespa_chunks should be called once with a single cleaned chunk
-    mock_batch_index.assert_called_once()
-    call_kwargs = mock_batch_index.call_args
-    indexed_chunks = call_kwargs.kwargs["chunks"]
-    assert len(indexed_chunks) == 1
-    assert indexed_chunks[0].source_document.id == "doc1"
-    assert call_kwargs.kwargs["index_name"] == "test_index"
-    assert call_kwargs.kwargs["multitenant"] is False
-
-
-@patch("onyx.document_index.vespa.vespa_document_index.batch_index_vespa_chunks")
-@patch("onyx.document_index.vespa.vespa_document_index.delete_vespa_chunks")
-@patch(
-    "onyx.document_index.vespa.vespa_document_index.get_document_chunk_ids",
-    return_value=[],
-)
-@patch("onyx.document_index.vespa.vespa_document_index._enrich_basic_chunk_info")
-def test_index_existing_doc_already_existed_true(
-    mock_enrich: MagicMock,
-    mock_get_chunk_ids: MagicMock,
-    mock_delete: MagicMock,
-    mock_batch_index: MagicMock,
-) -> None:
-    """Re-indexing a doc with previous chunks deletes old chunks, indexes
-    new ones, and returns already_existed=True."""
-    fake_chunk_ids = [uuid4(), uuid4()]
-    mock_enrich.return_value = _stub_enrich("doc1", old_chunk_cnt=5)
-    mock_get_chunk_ids.return_value = fake_chunk_ids
-
-    index = VespaDocumentIndex(
-        index_name="test_index",
-        tenant_state=TenantState(tenant_id="test_tenant", multitenant=False),
-        large_chunks_enabled=False,
-        httpx_client=MagicMock(),
-    )
-
-    chunk = _make_chunk("doc1")
-    metadata = _make_indexing_metadata(["doc1"], old_counts=[5], new_counts=[1])
-
-    results = index.index(chunks=[chunk], indexing_metadata=metadata)
-
-    assert len(results) == 1
-    assert results[0].already_existed is True
-
-    # Old chunks should be deleted
-    mock_delete.assert_called_once()
-    delete_kwargs = mock_delete.call_args.kwargs
-    assert delete_kwargs["doc_chunk_ids"] == fake_chunk_ids
-    assert delete_kwargs["index_name"] == "test_index"
-
-    # New chunk should be indexed
-    mock_batch_index.assert_called_once()
-    indexed_chunks = mock_batch_index.call_args.kwargs["chunks"]
-    assert len(indexed_chunks) == 1
-    assert indexed_chunks[0].source_document.id == "doc1"
-
-
-@patch("onyx.document_index.vespa.vespa_document_index.batch_index_vespa_chunks")
-@patch("onyx.document_index.vespa.vespa_document_index.delete_vespa_chunks")
-@patch(
-    "onyx.document_index.vespa.vespa_document_index.get_document_chunk_ids",
-    return_value=[],
-)
-@patch("onyx.document_index.vespa.vespa_document_index._enrich_basic_chunk_info")
-def test_index_multiple_docs(
-    mock_enrich: MagicMock,
-    mock_get_chunk_ids: MagicMock,  # noqa: ARG001
-    mock_delete: MagicMock,  # noqa: ARG001
-    mock_batch_index: MagicMock,
-) -> None:
-    """Indexing multiple documents returns one record per unique document."""
-    mock_enrich.side_effect = [
-        _stub_enrich("doc1", old_chunk_cnt=0),
-        _stub_enrich("doc2", old_chunk_cnt=3),
-    ]
-
-    index = VespaDocumentIndex(
-        index_name="test_index",
-        tenant_state=TenantState(tenant_id="test_tenant", multitenant=False),
-        large_chunks_enabled=False,
-        httpx_client=MagicMock(),
-    )
-
-    chunks = [
-        _make_chunk("doc1", chunk_id=0),
-        _make_chunk("doc1", chunk_id=1),
-        _make_chunk("doc2", chunk_id=0),
-    ]
-    metadata = _make_indexing_metadata(
-        ["doc1", "doc2"], old_counts=[0, 3], new_counts=[2, 1]
-    )
-
-    results = index.index(chunks=chunks, indexing_metadata=metadata)
-
-    result_map = {r.document_id: r.already_existed for r in results}
-    assert len(result_map) == 2
-    assert result_map["doc1"] is False
-    assert result_map["doc2"] is True
-
-    # All 3 chunks fit in one batch (BATCH_SIZE=128), so one call
-    mock_batch_index.assert_called_once()
-    indexed_chunks = mock_batch_index.call_args.kwargs["chunks"]
-    assert len(indexed_chunks) == 3
-    indexed_doc_ids = [c.source_document.id for c in indexed_chunks]
-    assert indexed_doc_ids == ["doc1", "doc1", "doc2"]
-
-
-@patch("onyx.document_index.vespa.vespa_document_index.batch_index_vespa_chunks")
-@patch("onyx.document_index.vespa.vespa_document_index.delete_vespa_chunks")
-@patch(
-    "onyx.document_index.vespa.vespa_document_index.get_document_chunk_ids",
-    return_value=[],
-)
-@patch("onyx.document_index.vespa.vespa_document_index._enrich_basic_chunk_info")
-def test_index_cleans_doc_ids(
-    mock_enrich: MagicMock,
-    mock_get_chunk_ids: MagicMock,  # noqa: ARG001
-    mock_delete: MagicMock,  # noqa: ARG001
-    mock_batch_index: MagicMock,
-) -> None:
-    """Documents with invalid Vespa characters get cleaned IDs, but
-    the returned DocumentInsertionRecord uses the original ID."""
-    doc_id_with_quote = "doc'1"
-    mock_enrich.return_value = _stub_enrich(doc_id_with_quote, old_chunk_cnt=0)
-
-    index = VespaDocumentIndex(
-        index_name="test_index",
-        tenant_state=TenantState(tenant_id="test_tenant", multitenant=False),
-        large_chunks_enabled=False,
-        httpx_client=MagicMock(),
-    )
-
-    chunk = _make_chunk(doc_id_with_quote)
-    metadata = _make_indexing_metadata(
-        [doc_id_with_quote], old_counts=[0], new_counts=[1]
-    )
-
-    results = index.index(chunks=[chunk], indexing_metadata=metadata)
-
-    assert len(results) == 1
-    # The returned ID should be the original (unclean) ID
-    assert results[0].document_id == doc_id_with_quote
-
-    # The chunk passed to batch_index_vespa_chunks should have the cleaned ID
-    indexed_chunks = mock_batch_index.call_args.kwargs["chunks"]
-    assert len(indexed_chunks) == 1
-    assert indexed_chunks[0].source_document.id == "doc_1"  # quote replaced with _
-
-
-@patch("onyx.document_index.vespa.vespa_document_index.batch_index_vespa_chunks")
-@patch("onyx.document_index.vespa.vespa_document_index.delete_vespa_chunks")
-@patch(
-    "onyx.document_index.vespa.vespa_document_index.get_document_chunk_ids",
-    return_value=[],
-)
-@patch("onyx.document_index.vespa.vespa_document_index._enrich_basic_chunk_info")
-def test_index_deduplicates_doc_ids_in_results(
-    mock_enrich: MagicMock,
-    mock_get_chunk_ids: MagicMock,  # noqa: ARG001
-    mock_delete: MagicMock,  # noqa: ARG001
-    mock_batch_index: MagicMock,
-) -> None:
-    """Multiple chunks from the same document produce only one
-    DocumentInsertionRecord."""
-    mock_enrich.return_value = _stub_enrich("doc1", old_chunk_cnt=0)
-
-    index = VespaDocumentIndex(
-        index_name="test_index",
-        tenant_state=TenantState(tenant_id="test_tenant", multitenant=False),
-        large_chunks_enabled=False,
-        httpx_client=MagicMock(),
-    )
-
-    chunks = [_make_chunk("doc1", chunk_id=i) for i in range(5)]
-    metadata = _make_indexing_metadata(["doc1"], old_counts=[0], new_counts=[5])
-
-    results = index.index(chunks=chunks, indexing_metadata=metadata)
-
-    assert len(results) == 1
-    assert results[0].document_id == "doc1"
-
-    # All 5 chunks should be passed to batch_index_vespa_chunks
-    mock_batch_index.assert_called_once()
-    indexed_chunks = mock_batch_index.call_args.kwargs["chunks"]
-    assert len(indexed_chunks) == 5
-    assert all(c.source_document.id == "doc1" for c in indexed_chunks)
-
-
-@patch("onyx.document_index.vespa.vespa_document_index.batch_index_vespa_chunks")
-@patch("onyx.document_index.vespa.vespa_document_index.delete_vespa_chunks")
-@patch(
-    "onyx.document_index.vespa.vespa_document_index.get_document_chunk_ids",
-    return_value=[],
-)
-@patch("onyx.document_index.vespa.vespa_document_index._enrich_basic_chunk_info")
-@patch(
-    "onyx.document_index.vespa.vespa_document_index.BATCH_SIZE",
-    3,
-)
-def test_index_respects_batch_size(
-    mock_enrich: MagicMock,
-    mock_get_chunk_ids: MagicMock,  # noqa: ARG001
-    mock_delete: MagicMock,  # noqa: ARG001
-    mock_batch_index: MagicMock,
-) -> None:
-    """When chunks exceed BATCH_SIZE, batch_index_vespa_chunks is called
-    multiple times with correctly sized batches."""
-    mock_enrich.return_value = _stub_enrich("doc1", old_chunk_cnt=0)
-
-    index = VespaDocumentIndex(
-        index_name="test_index",
-        tenant_state=TenantState(tenant_id="test_tenant", multitenant=False),
-        large_chunks_enabled=False,
-        httpx_client=MagicMock(),
-    )
-
-    chunks = [_make_chunk("doc1", chunk_id=i) for i in range(7)]
-    metadata = _make_indexing_metadata(["doc1"], old_counts=[0], new_counts=[7])
-
-    results = index.index(chunks=chunks, indexing_metadata=metadata)
-
-    assert len(results) == 1
-
-    # With BATCH_SIZE=3 and 7 chunks: batches of 3, 3, 1
-    assert mock_batch_index.call_count == 3
-    batch_sizes = [len(c.kwargs["chunks"]) for c in mock_batch_index.call_args_list]
-    assert batch_sizes == [3, 3, 1]
-
-    # Verify all chunks are accounted for and in order
-    all_indexed = [
-        chunk for c in mock_batch_index.call_args_list for chunk in c.kwargs["chunks"]
-    ]
-    assert len(all_indexed) == 7
-    assert [c.chunk_id for c in all_indexed] == list(range(7))
-
-
-@patch("onyx.document_index.vespa.vespa_document_index.batch_index_vespa_chunks")
-@patch("onyx.document_index.vespa.vespa_document_index.delete_vespa_chunks")
-@patch(
-    "onyx.document_index.vespa.vespa_document_index.get_document_chunk_ids",
-    return_value=[],
-)
-@patch("onyx.document_index.vespa.vespa_document_index._enrich_basic_chunk_info")
-def test_index_streams_chunks_lazily(
-    mock_enrich: MagicMock,
-    mock_get_chunk_ids: MagicMock,  # noqa: ARG001
-    mock_delete: MagicMock,  # noqa: ARG001
-    mock_batch_index: MagicMock,  # noqa: ARG001
-) -> None:
-    """Chunks are consumed lazily via a generator, not materialized upfront."""
-    mock_enrich.return_value = _stub_enrich("doc1", old_chunk_cnt=0)
-
-    index = VespaDocumentIndex(
-        index_name="test_index",
-        tenant_state=TenantState(tenant_id="test_tenant", multitenant=False),
-        large_chunks_enabled=False,
-        httpx_client=MagicMock(),
-    )
-
-    consumed: list[int] = []
-
-    def chunk_generator() -> Iterator[DocMetadataAwareIndexChunk]:
-        for i in range(3):
-            consumed.append(i)
-            yield _make_chunk("doc1", chunk_id=i)
-
-    metadata = _make_indexing_metadata(["doc1"], old_counts=[0], new_counts=[3])
-
-    # Before calling index, nothing consumed
-    gen = chunk_generator()
-    assert len(consumed) == 0
-
-    results = index.index(chunks=gen, indexing_metadata=metadata)
-
-    # After calling index, all chunks should have been consumed
-    assert consumed == [0, 1, 2]
-    assert len(results) == 1

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 +1,22 @@
+from typing import Any
+
+import pytest
+
+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, etc.
+
+            @property
+            def input_schema(self) -> dict[str, Any]:
+                return {}
+
+            @property
+            def output_schema(self) -> dict[str, Any]:
+                return {}

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 +1,60 @@
+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"]
+    assert "null" in props["user_email"]["type"]
+
+
+def test_output_schema_query_is_required() -> None:
+    schema = QueryProcessingSpec().output_schema
+    assert "query" in schema["required"]
+
+
+def test_output_schema_query_is_nullable() -> None:
+    # null means "reject the query"
+    props = QueryProcessingSpec().output_schema["properties"]
+    assert "null" in props["query"]["type"]
+
+
+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/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

deployment/docker_compose/install.sh

@@ -207,6 +207,16 @@ 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 +405,11 @@ 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 +426,11 @@ 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 +581,31 @@ 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"
@@ -745,6 +785,7 @@ 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
@@ -1084,6 +1125,25 @@ 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..."

examples/widget/package-lock.json

@@ -8,7 +8,7 @@
       "name": "widget",
       "version": "0.1.0",
       "dependencies": {
-        "next": "^16.1.5",
+        "next": "^16.1.7",
         "react": "^19",
         "react-dom": "^19",
         "react-markdown": "^10.1.0"
@@ -1023,9 +1023,9 @@
       }
     },
     "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": {
@@ -1039,9 +1039,9 @@
       }
     },
     "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"
       ],
@@ -1055,9 +1055,9 @@
       }
     },
     "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"
       ],
@@ -1071,9 +1071,9 @@
       }
     },
     "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"
       ],
@@ -1087,9 +1087,9 @@
       }
     },
     "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"
       ],
@@ -1103,9 +1103,9 @@
       }
     },
     "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"
       ],
@@ -1119,9 +1119,9 @@
       }
     },
     "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"
       ],
@@ -1135,9 +1135,9 @@
       }
     },
     "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"
       ],
@@ -1151,9 +1151,9 @@
       }
     },
     "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"
       ],
@@ -2564,12 +2564,15 @@
       "dev": true
     },
     "node_modules/baseline-browser-mapping": {
-      "version": "2.9.14",
-      "resolved": "https://registry.npmjs.org/baseline-browser-mapping/-/baseline-browser-mapping-2.9.14.tgz",
-      "integrity": "sha512-B0xUquLkiGLgHhpPBqvl7GWegWBUNuujQ6kXd/r1U38ElPT6Ok8KZ8e+FpUGEc2ZoRQUzq/aUnaKFc/svWUGSg==",
+      "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/brace-expansion": {
@@ -5926,14 +5929,14 @@
       "dev": true
     },
     "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"
@@ -5945,14 +5948,14 @@
         "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": {

examples/widget/package.json

@@ -9,7 +9,7 @@
     "lint": "next lint"
   },
   "dependencies": {
-    "next": "^16.1.5",
+    "next": "^16.1.7",
     "react": "^19",
     "react-dom": "^19",
     "react-markdown": "^10.1.0"

pyproject.toml

@@ -92,7 +92,7 @@ backend = [
     "python-gitlab==5.6.0",
     "python-pptx==0.6.23",
     "pypandoc_binary==1.16.2",
-    "pypdf==6.8.0",
+    "pypdf==6.9.1",
     "pytest-mock==3.12.0",
     "pytest-playwright==0.7.0",
     "python-docx==1.1.2",
@@ -245,6 +245,7 @@ select = [
   "ARG",
   "E",
   "F",
+  "S324",
   "W",
 ]
 

tools/ods/cmd/run-ci.go

@@ -17,6 +17,7 @@ import (
 type RunCIOptions struct {
 	DryRun bool
 	Yes    bool
+	Rerun  bool
 }
 
 // NewRunCICommand creates a new run-ci command
@@ -49,6 +50,7 @@ Example usage:
 
 	cmd.Flags().BoolVar(&opts.DryRun, "dry-run", false, "Perform all local operations but skip pushing to remote and creating PRs")
 	cmd.Flags().BoolVar(&opts.Yes, "yes", false, "Skip confirmation prompts and automatically proceed")
+	cmd.Flags().BoolVar(&opts.Rerun, "rerun", false, "Update an existing CI PR with the latest fork changes to re-trigger CI")
 
 	return cmd
 }
@@ -107,19 +109,44 @@ func runCI(cmd *cobra.Command, args []string, opts *RunCIOptions) {
 		log.Fatalf("PR #%s is not from a fork - CI should already run automatically", prNumber)
 	}
 
-	// Confirm before proceeding
-	if !opts.Yes {
-		if !prompt.Confirm(fmt.Sprintf("Create CI branch for PR #%s? (yes/no): ", prNumber)) {
-			log.Info("Exiting...")
-			return
-		}
-	}
-
 	// Create the CI branch
 	ciBranch := fmt.Sprintf("run-ci/%s", prNumber)
 	prTitle := fmt.Sprintf("chore: [Running GitHub actions for #%s]", prNumber)
 	prBody := fmt.Sprintf("This PR runs GitHub Actions CI for #%s.\n\n- [x] Override Linear Check\n\n**This PR should be closed (not merged) after CI completes.**", prNumber)
 
+	// Check if a CI PR already exists for this branch
+	existingPRURL, err := findExistingCIPR(ciBranch)
+	if err != nil {
+		log.Fatalf("Failed to check for existing CI PR: %v", err)
+	}
+
+	if existingPRURL != "" && !opts.Rerun {
+		log.Infof("A CI PR already exists for #%s: %s", prNumber, existingPRURL)
+		log.Info("Run with --rerun to update it with the latest fork changes and re-trigger CI.")
+		return
+	}
+
+	if opts.Rerun && existingPRURL == "" {
+		log.Warn("--rerun was specified but no existing open CI PR was found. A new PR will be created.")
+	}
+
+	if existingPRURL != "" && opts.Rerun {
+		log.Infof("Existing CI PR found: %s", existingPRURL)
+		log.Info("Will update the CI branch with the latest fork changes to re-trigger CI.")
+	}
+
+	// Confirm before proceeding
+	if !opts.Yes {
+		action := "Create CI branch"
+		if existingPRURL != "" {
+			action = "Update existing CI branch"
+		}
+		if !prompt.Confirm(fmt.Sprintf("%s for PR #%s? (yes/no): ", action, prNumber)) {
+			log.Info("Exiting...")
+			return
+		}
+	}
+
 	// Fetch the fork's branch
 	if forkRepo == "" {
 		log.Fatalf("Could not determine fork repository - headRepositoryOwner or headRepository.name is empty")
@@ -158,7 +185,11 @@ func runCI(cmd *cobra.Command, args []string, opts *RunCIOptions) {
 
 	if opts.DryRun {
 		log.Warnf("[DRY RUN] Would push CI branch: %s", ciBranch)
-		log.Warnf("[DRY RUN] Would create PR: %s", prTitle)
+		if existingPRURL == "" {
+			log.Warnf("[DRY RUN] Would create PR: %s", prTitle)
+		} else {
+			log.Warnf("[DRY RUN] Would update existing PR: %s", existingPRURL)
+		}
 		// Switch back to original branch
 		if err := git.RunCommand("switch", "--quiet", originalBranch); err != nil {
 			log.Warnf("Failed to switch back to original branch: %v", err)
@@ -176,6 +207,17 @@ func runCI(cmd *cobra.Command, args []string, opts *RunCIOptions) {
 		log.Fatalf("Failed to push CI branch: %v", err)
 	}
 
+	if existingPRURL != "" {
+		// PR already exists - force push is enough to re-trigger CI
+		log.Infof("Switching back to original branch: %s", originalBranch)
+		if err := git.RunCommand("switch", "--quiet", originalBranch); err != nil {
+			log.Warnf("Failed to switch back to original branch: %v", err)
+		}
+		log.Infof("CI PR updated successfully: %s", existingPRURL)
+		log.Info("The force push will re-trigger CI. Remember to close (not merge) this PR after CI completes!")
+		return
+	}
+
 	// Create PR using GitHub CLI
 	log.Info("Creating PR...")
 	prURL, err := createCIPR(ciBranch, prInfo.BaseRefName, prTitle, prBody)
@@ -217,6 +259,39 @@ func getPRInfo(prNumber string) (*PRInfo, error) {
 	return &prInfo, nil
 }
 
+// findExistingCIPR checks if an open PR already exists for the given CI branch.
+// Returns the PR URL if found, or empty string if not.
+func findExistingCIPR(headBranch string) (string, error) {
+	cmd := exec.Command("gh", "pr", "list",
+		"--head", headBranch,
+		"--state", "open",
+		"--json", "url",
+	)
+	output, err := cmd.Output()
+	if err != nil {
+		if exitErr, ok := err.(*exec.ExitError); ok {
+			return "", fmt.Errorf("%w: %s", err, string(exitErr.Stderr))
+		}
+		return "", err
+	}
+
+	var prs []struct {
+		URL string `json:"url"`
+	}
+	if err := json.Unmarshal(output, &prs); err != nil {
+		log.Debugf("Failed to parse PR list JSON: %v (raw: %s)", err, string(output))
+		return "", fmt.Errorf("failed to parse PR list: %w", err)
+	}
+
+	if len(prs) == 0 {
+		log.Debugf("No existing open PRs found for branch %s", headBranch)
+		return "", nil
+	}
+
+	log.Debugf("Found existing PR for branch %s: %s", headBranch, prs[0].URL)
+	return prs[0].URL, nil
+}
+
 // createCIPR creates a pull request for CI using the GitHub CLI
 func createCIPR(headBranch, baseBranch, title, body string) (string, error) {
 	cmd := exec.Command("gh", "pr", "create",

uv.lock

@@ -4481,7 +4481,7 @@ requires-dist = [
     { name = "pygithub", marker = "extra == 'backend'", specifier = "==2.5.0" },
     { name = "pympler", marker = "extra == 'backend'", specifier = "==1.1" },
     { name = "pypandoc-binary", marker = "extra == 'backend'", specifier = "==1.16.2" },
-    { name = "pypdf", marker = "extra == 'backend'", specifier = "==6.8.0" },
+    { name = "pypdf", marker = "extra == 'backend'", specifier = "==6.9.1" },
     { name = "pytest", marker = "extra == 'dev'", specifier = "==8.3.5" },
     { name = "pytest-alembic", marker = "extra == 'dev'", specifier = "==0.12.1" },
     { name = "pytest-asyncio", marker = "extra == 'dev'", specifier = "==1.3.0" },
@@ -5727,11 +5727,11 @@ wheels = [
 
 [[package]]
 name = "pypdf"
-version = "6.8.0"
+version = "6.9.1"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/b4/a3/e705b0805212b663a4c27b861c8a603dba0f8b4bb281f96f8e746576a50d/pypdf-6.8.0.tar.gz", hash = "sha256:cb7eaeaa4133ce76f762184069a854e03f4d9a08568f0e0623f7ea810407833b", size = 5307831, upload-time = "2026-03-09T13:37:40.591Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/f9/fb/dc2e8cb006e80b0020ed20d8649106fe4274e82d8e756ad3e24ade19c0df/pypdf-6.9.1.tar.gz", hash = "sha256:ae052407d33d34de0c86c5c729be6d51010bf36e03035a8f23ab449bca52377d", size = 5311551, upload-time = "2026-03-17T10:46:07.876Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/8c/ec/4ccf3bb86b1afe5d7176e1c8abcdbf22b53dd682ec2eda50e1caadcf6846/pypdf-6.8.0-py3-none-any.whl", hash = "sha256:2a025080a8dd73f48123c89c57174a5ff3806c71763ee4e49572dc90454943c7", size = 332177, upload-time = "2026-03-09T13:37:38.774Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/f4/75543fa802b86e72f87e9395440fe1a89a6d149887e3e55745715c3352ac/pypdf-6.9.1-py3-none-any.whl", hash = "sha256:f35a6a022348fae47e092a908339a8f3dc993510c026bb39a96718fc7185e89f", size = 333661, upload-time = "2026-03-17T10:46:06.286Z" },
 ]
 
 [[package]]

web/lib/opal/src/components/buttons/button/components.tsx

@@ -1,4 +1,3 @@
-import "@opal/components/buttons/button/styles.css";
 import "@opal/components/tooltip.css";
 import { Interactive, type InteractiveStatelessProps } from "@opal/core";
 import type { ContainerSizeVariants, ExtremaSizeVariants } from "@opal/types";
@@ -67,7 +66,7 @@ function Button({
   const labelEl = children ? (
     <span
       className={cn(
-        "opal-button-label",
+        "whitespace-nowrap",
         isLarge ? "font-main-ui-body " : "font-secondary-body",
         responsiveHideText && "hidden md:inline"
       )}
@@ -87,7 +86,7 @@ function Button({
           isLarge ? "default" : size === "2xs" ? "mini" : "compact"
         }
       >
-        <div className={cn("opal-button interactive-foreground")}>
+        <div className="flex flex-row items-center gap-1 interactive-foreground">
           {iconWrapper(Icon, size, !!children)}
 
           {labelEl}

web/lib/opal/src/components/buttons/button/styles.css

@@ -1,9 +0,0 @@
-/* Button — layout only; colors handled by Interactive.Stateless */
-
-.opal-button {
-  @apply flex flex-row items-center gap-1;
-}
-
-.opal-button-label {
-  @apply whitespace-nowrap;
-}

web/lib/opal/src/components/buttons/open-button/components.tsx

@@ -118,7 +118,7 @@ function OpenButton({
   const labelEl = children ? (
     <span
       className={cn(
-        "opal-button-label whitespace-nowrap",
+        "whitespace-nowrap",
         isLarge ? "font-main-ui-body" : "font-secondary-body"
       )}
     >
@@ -143,7 +143,7 @@ function OpenButton({
       >
         <div
           className={cn(
-            "opal-button interactive-foreground flex flex-row items-center",
+            "interactive-foreground flex flex-row items-center",
             justifyContent === "between" ? "w-full justify-between" : "gap-1",
             foldable &&
               justifyContent !== "between" &&

web/lib/opal/src/components/index.ts

@@ -52,3 +52,8 @@ export {
   type PaginationProps,
   type PaginationSize,
 } from "@opal/components/pagination/components";
+
+/* Table */
+export { Table } from "@opal/components/table/components";
+export { createTableColumns } from "@opal/components/table/columns";
+export type { DataTableProps } from "@opal/components/table/components";

web/lib/opal/src/components/table/ActionsContainer.tsx

@@ -1,5 +1,5 @@
-import { useTableSize } from "@/refresh-components/table/TableSizeContext";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+import { useTableSize } from "@opal/components/table/TableSizeContext";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
 
 interface ActionsContainerProps {
   type: "head" | "cell";

web/lib/opal/src/components/table/ColumnSortabilityPopover.tsx

@@ -7,11 +7,10 @@ import {
   type RowData,
   type SortingState,
 } from "@tanstack/react-table";
-import { Button } from "@opal/components";
+import { Button, LineItemButton } from "@opal/components";
 import { SvgArrowUpDown, SvgSortOrder, SvgCheck } from "@opal/icons";
 import Popover from "@/refresh-components/Popover";
 import Divider from "@/refresh-components/Divider";
-import LineItem from "@/refresh-components/buttons/LineItem";
 import Text from "@/refresh-components/texts/Text";
 
 // ---------------------------------------------------------------------------
@@ -21,7 +20,7 @@ import Text from "@/refresh-components/texts/Text";
 interface SortingPopoverProps<TData extends RowData = RowData> {
   table: Table<TData>;
   sorting: SortingState;
-  size?: "regular" | "small";
+  size?: "md" | "lg";
   footerText?: string;
   ascendingLabel?: string;
   descendingLabel?: string;
@@ -30,7 +29,7 @@ interface SortingPopoverProps<TData extends RowData = RowData> {
 function SortingPopover<TData extends RowData>({
   table,
   sorting,
-  size = "regular",
+  size = "lg",
   footerText,
   ascendingLabel = "Ascending",
   descendingLabel = "Descending",
@@ -48,8 +47,8 @@ function SortingPopover<TData extends RowData>({
         <Button
           icon={currentSort === null ? SvgArrowUpDown : SvgSortOrder}
           interaction={open ? "hover" : "rest"}
-          size={size === "small" ? "sm" : "md"}
-          prominence="internal"
+          size={size === "md" ? "sm" : "md"}
+          prominence="tertiary"
           tooltip="Sort"
         />
       </Popover.Trigger>
@@ -68,18 +67,20 @@ function SortingPopover<TData extends RowData>({
         >
           <Divider showTitle text="Sort by" />
 
-          <LineItem
-            selected={currentSort === null}
-            emphasized
+          <LineItemButton
+            selectVariant="select-heavy"
+            state={currentSort === null ? "selected" : "empty"}
+            title="Manual Ordering"
+            sizePreset="main-ui"
             rightChildren={
-              currentSort === null ? <SvgCheck size={16} /> : undefined
+              currentSort === null ? (
+                <SvgCheck size={16} className="text-action-link-05" />
+              ) : undefined
             }
             onClick={() => {
               table.resetSorting();
             }}
-          >
-            Manual Ordering
-          </LineItem>
+          />
 
           {sortableColumns.map((column) => {
             const isSorted = currentSort?.id === column.id;
@@ -89,11 +90,17 @@ function SortingPopover<TData extends RowData>({
                 : column.id;
 
             return (
-              <LineItem
+              <LineItemButton
                 key={column.id}
-                selected={isSorted}
-                emphasized
-                rightChildren={isSorted ? <SvgCheck size={16} /> : undefined}
+                selectVariant="select-heavy"
+                state={isSorted ? "selected" : "empty"}
+                title={label}
+                sizePreset="main-ui"
+                rightChildren={
+                  isSorted ? (
+                    <SvgCheck size={16} className="text-action-link-05" />
+                  ) : undefined
+                }
                 onClick={() => {
                   if (isSorted) {
                     table.resetSorting();
@@ -101,9 +108,7 @@ function SortingPopover<TData extends RowData>({
                   }
                   column.toggleSorting(false);
                 }}
-              >
-                {label}
-              </LineItem>
+              />
             );
           })}
 
@@ -111,31 +116,35 @@ function SortingPopover<TData extends RowData>({
             <>
               <Divider showTitle text="Sorting Order" />
 
-              <LineItem
-                selected={!currentSort.desc}
-                emphasized
+              <LineItemButton
+                selectVariant="select-heavy"
+                state={!currentSort.desc ? "selected" : "empty"}
+                title={ascendingLabel}
+                sizePreset="main-ui"
                 rightChildren={
-                  !currentSort.desc ? <SvgCheck size={16} /> : undefined
+                  !currentSort.desc ? (
+                    <SvgCheck size={16} className="text-action-link-05" />
+                  ) : undefined
                 }
                 onClick={() => {
                   table.setSorting([{ id: currentSort.id, desc: false }]);
                 }}
-              >
-                {ascendingLabel}
-              </LineItem>
+              />
 
-              <LineItem
-                selected={currentSort.desc}
-                emphasized
+              <LineItemButton
+                selectVariant="select-heavy"
+                state={currentSort.desc ? "selected" : "empty"}
+                title={descendingLabel}
+                sizePreset="main-ui"
                 rightChildren={
-                  currentSort.desc ? <SvgCheck size={16} /> : undefined
+                  currentSort.desc ? (
+                    <SvgCheck size={16} className="text-action-link-05" />
+                  ) : undefined
                 }
                 onClick={() => {
                   table.setSorting([{ id: currentSort.id, desc: true }]);
                 }}
-              >
-                {descendingLabel}
-              </LineItem>
+              />
             </>
           )}
         </Popover.Menu>
@@ -149,7 +158,7 @@ function SortingPopover<TData extends RowData>({
 // ---------------------------------------------------------------------------
 
 interface CreateSortingColumnOptions {
-  size?: "regular" | "small";
+  size?: "md" | "lg";
   footerText?: string;
   ascendingLabel?: string;
   descendingLabel?: string;

web/lib/opal/src/components/table/ColumnVisibilityPopover.tsx

@@ -7,10 +7,9 @@ import {
   type RowData,
   type VisibilityState,
 } from "@tanstack/react-table";
-import { Button } from "@opal/components";
+import { Button, LineItemButton, Tag } from "@opal/components";
 import { SvgColumn, SvgCheck } from "@opal/icons";
 import Popover from "@/refresh-components/Popover";
-import LineItem from "@/refresh-components/buttons/LineItem";
 import Divider from "@/refresh-components/Divider";
 
 // ---------------------------------------------------------------------------
@@ -20,18 +19,20 @@ import Divider from "@/refresh-components/Divider";
 interface ColumnVisibilityPopoverProps<TData extends RowData = RowData> {
   table: Table<TData>;
   columnVisibility: VisibilityState;
-  size?: "regular" | "small";
+  size?: "md" | "lg";
 }
 
 function ColumnVisibilityPopover<TData extends RowData>({
   table,
   columnVisibility,
-  size = "regular",
+  size = "lg",
 }: ColumnVisibilityPopoverProps<TData>) {
   const [open, setOpen] = useState(false);
-  const hideableColumns = table
+
+  // User-defined columns only (exclude internal qualifier/actions)
+  const dataColumns = table
     .getAllLeafColumns()
-    .filter((col) => col.getCanHide());
+    .filter((col) => !col.id.startsWith("__") && col.id !== "qualifier");
 
   return (
     <Popover open={open} onOpenChange={setOpen}>
@@ -39,8 +40,8 @@ function ColumnVisibilityPopover<TData extends RowData>({
         <Button
           icon={SvgColumn}
           interaction={open ? "hover" : "rest"}
-          size={size === "small" ? "sm" : "md"}
-          prominence="internal"
+          size={size === "md" ? "sm" : "md"}
+          prominence="tertiary"
           tooltip="Columns"
         />
       </Popover.Trigger>
@@ -48,7 +49,8 @@ function ColumnVisibilityPopover<TData extends RowData>({
       <Popover.Content width="lg" align="end" side="bottom">
         <Divider showTitle text="Shown Columns" />
         <Popover.Menu>
-          {hideableColumns.map((column) => {
+          {dataColumns.map((column) => {
+            const canHide = column.getCanHide();
             const isVisible = columnVisibility[column.id] !== false;
             const label =
               typeof column.columnDef.header === "string"
@@ -56,17 +58,23 @@ function ColumnVisibilityPopover<TData extends RowData>({
                 : column.id;
 
             return (
-              <LineItem
+              <LineItemButton
                 key={column.id}
-                selected={isVisible}
-                emphasized
-                rightChildren={isVisible ? <SvgCheck size={16} /> : undefined}
-                onClick={() => {
-                  column.toggleVisibility();
-                }}
-              >
-                {label}
-              </LineItem>
+                selectVariant="select-heavy"
+                state={isVisible ? "selected" : "empty"}
+                title={label}
+                sizePreset="main-ui"
+                rightChildren={
+                  !canHide ? (
+                    <div className="flex items-center">
+                      <Tag title="Always Shown" color="blue" />
+                    </div>
+                  ) : isVisible ? (
+                    <SvgCheck size={16} className="text-action-link-05" />
+                  ) : undefined
+                }
+                onClick={canHide ? () => column.toggleVisibility() : undefined}
+              />
             );
           })}
         </Popover.Menu>
@@ -80,7 +88,7 @@ function ColumnVisibilityPopover<TData extends RowData>({
 // ---------------------------------------------------------------------------
 
 interface CreateColumnVisibilityColumnOptions {
-  size?: "regular" | "small";
+  size?: "md" | "lg";
 }
 
 function createColumnVisibilityColumn<TData>(

web/lib/opal/src/components/table/DragOverlayRow.tsx

@@ -1,18 +1,17 @@
 import { memo } from "react";
 import { type Row, flexRender } from "@tanstack/react-table";
-import TableRow from "@/refresh-components/table/TableRow";
-import TableCell from "@/refresh-components/table/TableCell";
-import QualifierContainer from "@/refresh-components/table/QualifierContainer";
-import TableQualifier from "@/refresh-components/table/TableQualifier";
-import ActionsContainer from "@/refresh-components/table/ActionsContainer";
+import TableRow from "@opal/components/table/TableRow";
+import TableCell from "@opal/components/table/TableCell";
+import QualifierContainer from "@opal/components/table/QualifierContainer";
+import TableQualifier from "@opal/components/table/TableQualifier";
+import ActionsContainer from "@opal/components/table/ActionsContainer";
 import type {
   OnyxColumnDef,
   OnyxQualifierColumn,
-} from "@/refresh-components/table/types";
+} from "@opal/components/table/types";
 
 interface DragOverlayRowProps<TData> {
   row: Row<TData>;
-  variant?: "table" | "list";
   columnWidths?: Record<string, number>;
   columnKindMap?: Map<string, OnyxColumnDef<TData>>;
   qualifierColumn?: OnyxQualifierColumn<TData> | null;
@@ -21,7 +20,6 @@ interface DragOverlayRowProps<TData> {
 
 function DragOverlayRowInner<TData>({
   row,
-  variant,
   columnWidths,
   columnKindMap,
   qualifierColumn,
@@ -50,7 +48,7 @@ function DragOverlayRowInner<TData>({
         </colgroup>
       )}
       <tbody>
-        <TableRow variant={variant} selected={row.getIsSelected()}>
+        <TableRow selected={row.getIsSelected()}>
           {row.getVisibleCells().map((cell) => {
             const colDef = columnKindMap?.get(cell.column.id);
 

web/lib/opal/src/components/table/Footer.tsx

@@ -1,10 +1,10 @@
 "use client";
 
-import { cn } from "@/lib/utils";
-import { Button, Pagination } from "@opal/components";
+import { cn } from "@opal/utils";
+import { Button, Pagination, SelectButton } from "@opal/components";
 import Text from "@/refresh-components/texts/Text";
-import { useTableSize } from "@/refresh-components/table/TableSizeContext";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+import { useTableSize } from "@opal/components/table/TableSizeContext";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
 import { SvgEye, SvgXCircle } from "@opal/icons";
 import type { ReactNode } from "react";
 
@@ -27,9 +27,11 @@ interface FooterSelectionModeProps {
   selectionState: SelectionState;
   /** Number of currently selected items. */
   selectedCount: number;
-  /** If provided, renders a "View" icon button when items are selected. */
+  /** Toggle view-filter on/off. */
   onView?: () => void;
-  /** If provided, renders a "Clear" icon button when items are selected. */
+  /** Whether the view-filter is currently active. */
+  isViewingSelected?: boolean;
+  /** Clears all selections. */
   onClear?: () => void;
   /** Number of items displayed per page. */
   pageSize: number;
@@ -41,7 +43,9 @@ interface FooterSelectionModeProps {
   totalPages: number;
   /** Called when the user navigates to a different page. */
   onPageChange: (page: number) => void;
-  /** Controls overall footer sizing. `"regular"` (default) or `"small"`. */
+  /** Unit label for count pagination. @default "items" */
+  units?: string;
+  /** Controls overall footer sizing. `"lg"` (default) or `"md"`. */
   size?: TableSize;
   className?: string;
 }
@@ -67,6 +71,8 @@ interface FooterSummaryModeProps {
   onPageChange: (page: number) => void;
   /** Optional extra element rendered after the summary text (e.g. a download icon). */
   leftExtra?: ReactNode;
+  /** Unit label for the summary text, e.g. "users". */
+  units?: string;
   className?: string;
 }
 
@@ -84,9 +90,10 @@ export type FooterProps = FooterSelectionModeProps | FooterSummaryModeProps;
 function getSelectionMessage(
   state: SelectionState,
   multi: boolean,
-  count: number
+  count: number,
+  isViewingSelected: boolean
 ): string {
-  if (state === "none") {
+  if (state === "none" && !isViewingSelected) {
     return multi ? "Select items to continue" : "Select an item to continue";
   }
   if (!multi) return "Item selected";
@@ -100,7 +107,7 @@ function getSelectionMessage(
  */
 export default function Footer(props: FooterProps) {
   const resolvedSize = useTableSize();
-  const isSmall = resolvedSize === "small";
+  const isSmall = resolvedSize === "md";
   return (
     <div
       className={cn(
@@ -118,6 +125,7 @@ export default function Footer(props: FooterProps) {
             multiSelect={props.multiSelect}
             selectedCount={props.selectedCount}
             onView={props.onView}
+            isViewingSelected={props.isViewingSelected}
             onClear={props.onClear}
             isSmall={isSmall}
           />
@@ -127,6 +135,7 @@ export default function Footer(props: FooterProps) {
               rangeStart={props.rangeStart}
               rangeEnd={props.rangeEnd}
               totalItems={props.totalItems}
+              units={props.units}
               isSmall={isSmall}
             />
             {props.leftExtra}
@@ -144,7 +153,7 @@ export default function Footer(props: FooterProps) {
             currentPage={props.currentPage}
             totalPages={props.totalPages}
             onChange={props.onPageChange}
-            units="items"
+            units={props.units}
             size={isSmall ? "sm" : "md"}
           />
         ) : (
@@ -169,6 +178,7 @@ interface SelectionLeftProps {
   multiSelect: boolean;
   selectedCount: number;
   onView?: () => void;
+  isViewingSelected?: boolean;
   onClear?: () => void;
   isSmall: boolean;
 }
@@ -178,15 +188,19 @@ function SelectionLeft({
   multiSelect,
   selectedCount,
   onView,
+  isViewingSelected = false,
   onClear,
   isSmall,
 }: SelectionLeftProps) {
   const message = getSelectionMessage(
     selectionState,
     multiSelect,
-    selectedCount
+    selectedCount,
+    isViewingSelected
   );
   const hasSelection = selectionState !== "none";
+  // Show buttons when items are selected OR when the view filter is active
+  const showActions = hasSelection || isViewingSelected;
 
   return (
     <div className="flex flex-row gap-1 items-center justify-center w-fit flex-shrink-0 h-fit px-1">
@@ -204,22 +218,22 @@ function SelectionLeft({
         </Text>
       )}
 
-      {hasSelection && (
+      {showActions && (
         <div className="flex flex-row items-center w-fit flex-shrink-0 h-fit">
           {onView && (
-            <Button
+            <SelectButton
               icon={SvgEye}
+              state={isViewingSelected ? "selected" : "empty"}
               onClick={onView}
-              tooltip="View"
+              tooltip="View selected"
               size={isSmall ? "sm" : "md"}
-              prominence="tertiary"
             />
           )}
           {onClear && (
             <Button
               icon={SvgXCircle}
               onClick={onClear}
-              tooltip="Clear selection"
+              tooltip="Deselect all"
               size={isSmall ? "sm" : "md"}
               prominence="tertiary"
             />
@@ -234,6 +248,7 @@ interface SummaryLeftProps {
   rangeStart: number;
   rangeEnd: number;
   totalItems: number;
+  units?: string;
   isSmall: boolean;
 }
 
@@ -241,8 +256,10 @@ function SummaryLeft({
   rangeStart,
   rangeEnd,
   totalItems,
+  units,
   isSmall,
 }: SummaryLeftProps) {
+  const suffix = units ? ` ${units}` : "";
   return (
     <div className="flex flex-row gap-1 items-center w-fit h-fit px-1">
       {isSmall ? (
@@ -255,6 +272,7 @@ function SummaryLeft({
           <Text as="span" secondaryMono text03>
             {totalItems}
           </Text>
+          {suffix}
         </Text>
       ) : (
         <Text mainUiMuted text03>
@@ -266,6 +284,7 @@ function SummaryLeft({
           <Text as="span" mainUiMono text03>
             {totalItems}
           </Text>
+          {suffix}
         </Text>
       )}
     </div>

web/lib/opal/src/components/table/QualifierContainer.tsx

@@ -1,5 +1,5 @@
-import { useTableSize } from "@/refresh-components/table/TableSizeContext";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+import { useTableSize } from "@opal/components/table/TableSizeContext";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
 
 interface QualifierContainerProps {
   type: "head" | "cell";

web/lib/opal/src/components/table/README.md

@@ -0,0 +1,82 @@
+# Table
+
+Config-driven table component with sorting, pagination, column visibility,
+row selection, drag-and-drop reordering, and server-side mode.
+
+## Usage
+
+```tsx
+import { Table, createTableColumns } from "@opal/components";
+
+interface User {
+  id: string;
+  email: string;
+  name: string | null;
+  status: "active" | "invited";
+}
+
+const tc = createTableColumns<User>();
+
+const columns = [
+  tc.qualifier({ content: "avatar-user", getInitials: (r) => r.name?.[0] ?? "?" }),
+  tc.column("email", {
+    header: "Name",
+    weight: 22,
+    minWidth: 140,
+    cell: (email, row) => <span>{row.name ?? email}</span>,
+  }),
+  tc.column("status", {
+    header: "Status",
+    weight: 14,
+    cell: (status) => <span>{status}</span>,
+  }),
+  tc.actions(),
+];
+
+function UsersTable({ users }: { users: User[] }) {
+  return (
+    <Table
+      data={users}
+      columns={columns}
+      getRowId={(r) => r.id}
+      pageSize={10}
+      footer={{ mode: "summary" }}
+    />
+  );
+}
+```
+
+## Props
+
+| Prop | Type | Default | Description |
+|------|------|---------|-------------|
+| `data` | `TData[]` | required | Row data array |
+| `columns` | `OnyxColumnDef<TData>[]` | required | Column definitions from `createTableColumns()` |
+| `getRowId` | `(row: TData) => string` | required | Unique row identifier |
+| `pageSize` | `number` | `10` | Rows per page (`Infinity` disables pagination) |
+| `size` | `"md" \| "lg"` | `"lg"` | Density variant |
+| `footer` | `DataTableFooterConfig` | — | Footer mode (`"selection"` or `"summary"`) |
+| `initialSorting` | `SortingState` | — | Initial sort state |
+| `initialColumnVisibility` | `VisibilityState` | — | Initial column visibility |
+| `draggable` | `DataTableDraggableConfig` | — | Enable drag-and-drop reordering |
+| `onSelectionChange` | `(ids: string[]) => void` | — | Selection callback |
+| `onRowClick` | `(row: TData) => void` | — | Row click handler |
+| `searchTerm` | `string` | — | Global text filter |
+| `height` | `number \| string` | — | Max scrollable height |
+| `headerBackground` | `string` | — | Sticky header background |
+| `serverSide` | `ServerSideConfig` | — | Server-side pagination/sorting/filtering |
+| `emptyState` | `ReactNode` | — | Empty state content |
+
+## Column Builder
+
+`createTableColumns<TData>()` returns a builder with:
+
+- `tc.qualifier(opts)` — leading avatar/icon/checkbox column
+- `tc.column(accessor, opts)` — data column with sorting/resizing
+- `tc.displayColumn(opts)` — non-accessor custom column
+- `tc.actions(opts)` — trailing actions column with visibility/sorting popovers
+
+## Footer Modes
+
+- **`"selection"`** — shows selection count, optional view/clear buttons, count pagination
+- **`"summary"`** — shows "Showing X~Y of Z", list pagination, optional extra element

web/lib/opal/src/components/table/Table.stories.tsx

@@ -0,0 +1,148 @@
+import type { Meta, StoryObj } from "@storybook/react";
+import { Table, createTableColumns } from "@opal/components";
+
+// ---------------------------------------------------------------------------
+// Sample data
+// ---------------------------------------------------------------------------
+
+interface User {
+  id: string;
+  email: string;
+  name: string;
+  role: "admin" | "user" | "viewer";
+  status: "active" | "invited" | "inactive";
+}
+
+const USERS: User[] = [
+  {
+    id: "1",
+    email: "alice@example.com",
+    name: "Alice Johnson",
+    role: "admin",
+    status: "active",
+  },
+  {
+    id: "2",
+    email: "bob@example.com",
+    name: "Bob Smith",
+    role: "user",
+    status: "active",
+  },
+  {
+    id: "3",
+    email: "carol@example.com",
+    name: "Carol White",
+    role: "viewer",
+    status: "invited",
+  },
+  {
+    id: "4",
+    email: "dave@example.com",
+    name: "Dave Brown",
+    role: "user",
+    status: "inactive",
+  },
+  {
+    id: "5",
+    email: "eve@example.com",
+    name: "Eve Davis",
+    role: "admin",
+    status: "active",
+  },
+  {
+    id: "6",
+    email: "frank@example.com",
+    name: "Frank Miller",
+    role: "viewer",
+    status: "active",
+  },
+  {
+    id: "7",
+    email: "grace@example.com",
+    name: "Grace Lee",
+    role: "user",
+    status: "invited",
+  },
+  {
+    id: "8",
+    email: "hank@example.com",
+    name: "Hank Wilson",
+    role: "user",
+    status: "active",
+  },
+  {
+    id: "9",
+    email: "iris@example.com",
+    name: "Iris Taylor",
+    role: "viewer",
+    status: "active",
+  },
+  {
+    id: "10",
+    email: "jack@example.com",
+    name: "Jack Moore",
+    role: "admin",
+    status: "active",
+  },
+  {
+    id: "11",
+    email: "kate@example.com",
+    name: "Kate Anderson",
+    role: "user",
+    status: "inactive",
+  },
+  {
+    id: "12",
+    email: "leo@example.com",
+    name: "Leo Thomas",
+    role: "viewer",
+    status: "active",
+  },
+];
+
+// ---------------------------------------------------------------------------
+// Columns
+// ---------------------------------------------------------------------------
+
+const tc = createTableColumns<User>();
+
+const columns = [
+  tc.qualifier({
+    content: "avatar-user",
+    getInitials: (r) =>
+      r.name
+        .split(" ")
+        .map((n) => n[0])
+        .join(""),
+  }),
+  tc.column("name", { header: "Name", weight: 25, minWidth: 120 }),
+  tc.column("email", { header: "Email", weight: 30, minWidth: 160 }),
+  tc.column("role", { header: "Role", weight: 15, minWidth: 80 }),
+  tc.column("status", { header: "Status", weight: 15, minWidth: 80 }),
+  tc.actions(),
+];
+
+// ---------------------------------------------------------------------------
+// Story
+// ---------------------------------------------------------------------------
+
+const meta: Meta<typeof Table> = {
+  title: "opal/components/Table",
+  component: Table,
+  tags: ["autodocs"],
+};
+
+export default meta;
+type Story = StoryObj<typeof Table>;
+
+export const Default: Story = {
+  render: () => (
+    <Table
+      data={USERS}
+      columns={columns}
+      getRowId={(r) => r.id}
+      pageSize={8}
+      footer={{ mode: "summary" }}
+    />
+  ),
+};

web/lib/opal/src/components/table/TableCell.tsx

@@ -1,6 +1,6 @@
-import { cn } from "@/lib/utils";
-import { useTableSize } from "@/refresh-components/table/TableSizeContext";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+import { cn } from "@opal/utils";
+import { useTableSize } from "@opal/components/table/TableSizeContext";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
 import type { WithoutStyles } from "@/types";
 
 interface TableCellProps

web/lib/opal/src/components/table/TableElement.tsx

@@ -0,0 +1,71 @@
+import React from "react";
+import { cn } from "@opal/utils";
+import type { WithoutStyles } from "@/types";
+import type { ExtremaSizeVariants, SizeVariants } from "@opal/types";
+
+// ---------------------------------------------------------------------------
+// Types
+// ---------------------------------------------------------------------------
+
+type TableSize = Extract<SizeVariants, "md" | "lg">;
+type TableVariant = "rows" | "cards";
+type TableQualifier = "simple" | "avatar" | "icon";
+type SelectionBehavior = "no-select" | "single-select" | "multi-select";
+
+interface TableProps
+  extends WithoutStyles<React.TableHTMLAttributes<HTMLTableElement>> {
+  ref?: React.Ref<HTMLTableElement>;
+  /** Size preset for the table. @default "lg" */
+  size?: TableSize;
+  /** Visual row variant. @default "cards" */
+  variant?: TableVariant;
+  /** Row selection behavior. @default "no-select" */
+  selectionBehavior?: SelectionBehavior;
+  /** Leading qualifier column type. @default null */
+  qualifier?: TableQualifier;
+  /** Height behavior. `"fit"` = shrink to content, `"full"` = fill available space. */
+  heightVariant?: ExtremaSizeVariants;
+  /** Explicit pixel width for the table (e.g. from `table.getTotalSize()`).
+   *  When provided the table uses exactly this width instead of stretching
+   *  to fill its container, which prevents `table-layout: fixed` from
+   *  redistributing extra space across columns on resize. */
+  width?: number;
+}
+
+// ---------------------------------------------------------------------------
+// Component
+// ---------------------------------------------------------------------------
+
+function Table({
+  ref,
+  size = "lg",
+  variant = "cards",
+  selectionBehavior = "no-select",
+  qualifier = "simple",
+  heightVariant,
+  width,
+  ...props
+}: TableProps) {
+  return (
+    <table
+      ref={ref}
+      className={cn("border-separate border-spacing-0", !width && "min-w-full")}
+      style={{ tableLayout: "fixed", width }}
+      data-size={size}
+      data-variant={variant}
+      data-selection={selectionBehavior}
+      data-qualifier={qualifier}
+      data-height={heightVariant}
+      {...props}
+    />
+  );
+}
+
+export default Table;
+export type {
+  TableProps,
+  TableSize,
+  TableVariant,
+  TableQualifier,
+  SelectionBehavior,
+};

web/lib/opal/src/components/table/TableHead.tsx

@@ -1,7 +1,7 @@
-import { cn } from "@/lib/utils";
+import { cn } from "@opal/utils";
 import Text from "@/refresh-components/texts/Text";
-import { useTableSize } from "@/refresh-components/table/TableSizeContext";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+import { useTableSize } from "@opal/components/table/TableSizeContext";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
 import type { WithoutStyles } from "@/types";
 import { Button } from "@opal/components";
 import { SvgChevronDown, SvgChevronUp, SvgHandle, SvgSort } from "@opal/icons";
@@ -30,7 +30,7 @@ interface TableHeadCustomProps {
   icon?: (sorted: SortDirection) => IconFunctionComponent;
   /** Text alignment for the column. Defaults to `"left"`. */
   alignment?: "left" | "center" | "right";
-  /** Cell density. `"small"` uses tighter padding for denser layouts. */
+  /** Cell density. `"md"` uses tighter padding for denser layouts. */
   size?: TableSize;
   /** Column width in pixels. Applied as an inline style on the `<th>`. */
   width?: number;
@@ -88,7 +88,7 @@ export default function TableHead({
 }: TableHeadProps) {
   const contextSize = useTableSize();
   const resolvedSize = size ?? contextSize;
-  const isSmall = resolvedSize === "small";
+  const isSmall = resolvedSize === "md";
   return (
     <th
       {...thProps}

web/lib/opal/src/components/table/TableQualifier.tsx

@@ -1,12 +1,12 @@
 "use client";
 
 import React from "react";
-import { cn } from "@/lib/utils";
-import { useTableSize } from "@/refresh-components/table/TableSizeContext";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+import { cn } from "@opal/utils";
+import { useTableSize } from "@opal/components/table/TableSizeContext";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
 import { SvgUser } from "@opal/icons";
 import type { IconFunctionComponent } from "@opal/types";
-import type { QualifierContentType } from "@/refresh-components/table/types";
+import type { QualifierContentType } from "@opal/components/table/types";
 import Checkbox from "@/refresh-components/inputs/Checkbox";
 import Text from "@/refresh-components/texts/Text";
 
@@ -35,8 +35,8 @@ interface TableQualifierProps {
 }
 
 const iconSizes = {
-  regular: 16,
-  small: 14,
+  lg: 16,
+  md: 14,
 } as const;
 
 function getQualifierStyles(selected: boolean, disabled: boolean) {
@@ -62,9 +62,9 @@ function getQualifierStyles(selected: boolean, disabled: boolean) {
     container: "bg-background-tint-01",
     icon: "stroke-text-03",
     overlay:
-      "flex opacity-0 group-hover:opacity-100 group-focus-within:opacity-100 bg-background-tint-01",
+      "flex opacity-0 group-hover/row:opacity-100 group-focus-within/row:opacity-100 bg-background-tint-01",
     overlayImage:
-      "flex opacity-0 group-hover:opacity-100 group-focus-within:opacity-100 bg-mask-01 group-hover:backdrop-blur-02 group-focus-within:backdrop-blur-02",
+      "flex opacity-0 group-hover/row:opacity-100 group-focus-within/row:opacity-100 bg-mask-01 group-hover/row:backdrop-blur-02 group-focus-within/row:backdrop-blur-02",
   };
 }
 
@@ -115,7 +115,7 @@ function TableQualifier({
           <div
             className={cn(
               "flex items-center justify-center rounded-full bg-background-neutral-inverted-00",
-              resolvedSize === "regular" ? "h-7 w-7" : "h-6 w-6"
+              resolvedSize === "lg" ? "h-7 w-7" : "h-6 w-6"
             )}
           >
             <Text
@@ -138,30 +138,36 @@ function TableQualifier({
     <div
       className={cn(
         "group relative inline-flex shrink-0 items-center justify-center",
-        resolvedSize === "regular" ? "h-9 w-9" : "h-7 w-7",
+        resolvedSize === "lg" ? "h-9 w-9" : "h-7 w-7",
         disabled ? "cursor-not-allowed" : "cursor-default",
         className
       )}
     >
-      {/* Inner qualifier container */}
-      <div
-        className={cn(
-          "flex items-center justify-center overflow-hidden transition-colors",
-          resolvedSize === "regular" ? "h-9 w-9" : "h-7 w-7",
-          isRound ? "rounded-full" : "rounded-08",
-          styles.container,
-          content === "image" && disabled && !selected && "opacity-50"
-        )}
-      >
-        {renderContent()}
-      </div>
+      {/* Inner qualifier container — no background for "simple" */}
+      {content !== "simple" && (
+        <div
+          className={cn(
+            "flex items-center justify-center overflow-hidden transition-colors",
+            resolvedSize === "lg" ? "h-9 w-9" : "h-7 w-7",
+            isRound ? "rounded-full" : "rounded-08",
+            styles.container,
+            content === "image" && disabled && !selected && "opacity-50"
+          )}
+        >
+          {renderContent()}
+        </div>
+      )}
 
       {/* Selection overlay */}
       {selectable && (
         <div
           className={cn(
             "absolute inset-0 items-center justify-center",
-            isRound ? "rounded-full" : "rounded-08",
+            content === "simple"
+              ? "flex"
+              : isRound
+                ? "rounded-full"
+                : "rounded-08",
             content === "simple"
               ? "flex"
               : content === "image"

web/lib/opal/src/components/table/TableRow.tsx

@@ -1,8 +1,8 @@
 "use client";
 
-import { cn } from "@/lib/utils";
-import { useTableSize } from "@/refresh-components/table/TableSizeContext";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+import { cn } from "@opal/utils";
+import { useTableSize } from "@opal/components/table/TableSizeContext";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
 import type { WithoutStyles } from "@/types";
 import { useSortable } from "@dnd-kit/sortable";
 import { CSS } from "@dnd-kit/utilities";
@@ -18,8 +18,6 @@ interface TableRowProps
   selected?: boolean;
   /** Disables interaction and applies disabled styling */
   disabled?: boolean;
-  /** Visual variant: "table" adds a bottom border, "list" adds rounded corners. Defaults to "list". */
-  variant?: "table" | "list";
   /** When provided, makes this row sortable via @dnd-kit */
   sortableId?: string;
   /** Show drag handle overlay. Defaults to true when sortableId is set. */
@@ -36,7 +34,6 @@ function SortableTableRow({
   sortableId,
   showDragHandle = true,
   size,
-  variant = "list",
   selected,
   disabled,
   ref: _externalRef,
@@ -66,7 +63,6 @@ function SortableTableRow({
       ref={setNodeRef}
       style={style}
       className="tbl-row group/row"
-      data-variant={variant}
       data-drag-handle={showDragHandle || undefined}
       data-selected={selected || undefined}
       data-disabled={disabled || undefined}
@@ -95,7 +91,7 @@ function SortableTableRow({
             {...listeners}
           >
             <SvgHandle
-              size={resolvedSize === "small" ? 12 : 16}
+              size={resolvedSize === "md" ? 12 : 16}
               className="text-border-02"
             />
           </button>
@@ -113,7 +109,6 @@ function TableRow({
   sortableId,
   showDragHandle,
   size,
-  variant = "list",
   selected,
   disabled,
   ref,
@@ -125,7 +120,6 @@ function TableRow({
         sortableId={sortableId}
         showDragHandle={showDragHandle}
         size={size}
-        variant={variant}
         selected={selected}
         disabled={disabled}
         ref={ref}
@@ -138,7 +132,6 @@ function TableRow({
     <tr
       ref={ref}
       className="tbl-row group/row"
-      data-variant={variant}
       data-selected={selected || undefined}
       data-disabled={disabled || undefined}
       {...props}

web/lib/opal/src/components/table/TableSizeContext.tsx

@@ -1,10 +1,11 @@
 "use client";
 
 import { createContext, useContext } from "react";
+import type { SizeVariants } from "@opal/types";
 
-type TableSize = "regular" | "small";
+type TableSize = Extract<SizeVariants, "md" | "lg">;
 
-const TableSizeContext = createContext<TableSize>("regular");
+const TableSizeContext = createContext<TableSize>("lg");
 
 interface TableSizeProviderProps {
   size: TableSize;

web/lib/opal/src/components/table/columns.ts

@@ -13,10 +13,10 @@ import type {
   OnyxDataColumn,
   OnyxDisplayColumn,
   OnyxActionsColumn,
-} from "@/refresh-components/table/types";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+} from "@opal/components/table/types";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
 import type { IconFunctionComponent } from "@opal/types";
-import type { SortDirection } from "@/refresh-components/table/TableHead";
+import type { SortDirection } from "@opal/components/table/TableHead";
 
 // ---------------------------------------------------------------------------
 // Qualifier column config
@@ -160,7 +160,7 @@ export function createTableColumns<TData>(): TableColumnsBuilder<TData> {
         id: "qualifier",
         def,
         width: (size: TableSize) =>
-          size === "small" ? { fixed: 40 } : { fixed: 56 },
+          size === "md" ? { fixed: 36 } : { fixed: 44 },
         content,
         headerContentType: config?.headerContentType,
         getInitials: config?.getInitials,
@@ -241,14 +241,29 @@ export function createTableColumns<TData>(): TableColumnsBuilder<TData> {
           : () => null,
       };
 
+      const showVisibility = config?.showColumnVisibility ?? true;
+      const showSorting = config?.showSorting ?? true;
+      const buttonCount = (showVisibility ? 1 : 0) + (showSorting ? 1 : 0);
+
+      // Icon button sizes: "md" button = 28px, "sm" button = 24px
+      // px-1 on .tbl-actions = 4px each side = 8px total
+      const BUTTON_MD = 28;
+      const BUTTON_SM = 24;
+      const PADDING = 8;
+
       return {
         kind: "actions",
         id: "__actions",
         def,
-        width: (size: TableSize) =>
-          size === "small" ? { fixed: 20 } : { fixed: 88 },
-        showColumnVisibility: config?.showColumnVisibility ?? true,
-        showSorting: config?.showSorting ?? true,
+        width: (size: TableSize) => ({
+          fixed:
+            Math.max(
+              buttonCount * (size === "md" ? BUTTON_SM : BUTTON_MD),
+              size === "md" ? BUTTON_SM : BUTTON_MD
+            ) + PADDING,
+        }),
+        showColumnVisibility: showVisibility,
+        showSorting: showSorting,
         sortingFooterText: config?.sortingFooterText,
       };
     },

web/lib/opal/src/components/table/components.tsx

@@ -1,39 +1,56 @@
 "use client";
 "use no memo";
 
+import "@opal/components/table/styles.css";
+
 import { useEffect, useMemo } from "react";
 import { flexRender } from "@tanstack/react-table";
 import useDataTable, {
   toOnyxSortDirection,
-} from "@/refresh-components/table/hooks/useDataTable";
-import useColumnWidths from "@/refresh-components/table/hooks/useColumnWidths";
-import useDraggableRows from "@/refresh-components/table/hooks/useDraggableRows";
-import Table from "@/refresh-components/table/Table";
-import TableHeader from "@/refresh-components/table/TableHeader";
-import TableBody from "@/refresh-components/table/TableBody";
-import TableRow from "@/refresh-components/table/TableRow";
-import TableHead from "@/refresh-components/table/TableHead";
-import TableCell from "@/refresh-components/table/TableCell";
-import TableQualifier from "@/refresh-components/table/TableQualifier";
-import QualifierContainer from "@/refresh-components/table/QualifierContainer";
-import ActionsContainer from "@/refresh-components/table/ActionsContainer";
-import DragOverlayRow from "@/refresh-components/table/DragOverlayRow";
-import Footer from "@/refresh-components/table/Footer";
-import { TableSizeProvider } from "@/refresh-components/table/TableSizeContext";
-import { ColumnVisibilityPopover } from "@/refresh-components/table/ColumnVisibilityPopover";
-import { SortingPopover } from "@/refresh-components/table/SortingPopover";
-import type { WidthConfig } from "@/refresh-components/table/hooks/useColumnWidths";
+} from "@opal/components/table/hooks/useDataTable";
+import useColumnWidths from "@opal/components/table/hooks/useColumnWidths";
+import useDraggableRows from "@opal/components/table/hooks/useDraggableRows";
+import TableElement from "@opal/components/table/TableElement";
+import TableHeader from "@opal/components/table/TableHeader";
+import TableBody from "@opal/components/table/TableBody";
+import TableRow from "@opal/components/table/TableRow";
+import TableHead from "@opal/components/table/TableHead";
+import TableCell from "@opal/components/table/TableCell";
+import TableQualifier from "@opal/components/table/TableQualifier";
+import QualifierContainer from "@opal/components/table/QualifierContainer";
+import ActionsContainer from "@opal/components/table/ActionsContainer";
+import DragOverlayRow from "@opal/components/table/DragOverlayRow";
+import Footer from "@opal/components/table/Footer";
+import Checkbox from "@/refresh-components/inputs/Checkbox";
+import { TableSizeProvider } from "@opal/components/table/TableSizeContext";
+import { ColumnVisibilityPopover } from "@opal/components/table/ColumnVisibilityPopover";
+import { SortingPopover } from "@opal/components/table/ColumnSortabilityPopover";
+import type { WidthConfig } from "@opal/components/table/hooks/useColumnWidths";
 import type { ColumnDef } from "@tanstack/react-table";
-import { cn } from "@/lib/utils";
+import { cn } from "@opal/utils";
 import type {
-  DataTableProps,
+  DataTableProps as BaseDataTableProps,
   DataTableFooterConfig,
   OnyxColumnDef,
   OnyxDataColumn,
   OnyxQualifierColumn,
   OnyxActionsColumn,
-} from "@/refresh-components/table/types";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+} from "@opal/components/table/types";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
+
+// ---------------------------------------------------------------------------
+// Qualifier × SelectionBehavior
+// ---------------------------------------------------------------------------
+
+type Qualifier = "simple" | "avatar" | "icon";
+type SelectionBehavior = "no-select" | "single-select" | "multi-select";
+
+export type DataTableProps<TData> = BaseDataTableProps<TData> & {
+  /** Leading qualifier column type. @default "simple" */
+  qualifier?: Qualifier;
+  /** Row selection behavior. @default "no-select" */
+  selectionBehavior?: SelectionBehavior;
+};
 
 // ---------------------------------------------------------------------------
 // Internal: resolve size-dependent widths and build TanStack columns
@@ -57,6 +74,7 @@ function processColumns<TData>(
   const columnMinWidths: Record<string, number> = {};
   const columnKindMap = new Map<string, OnyxColumnDef<TData>>();
   let qualifierColumn: OnyxQualifierColumn<TData> | null = null;
+  let firstDataColumnSeen = false;
 
   for (const col of columns) {
     const resolvedWidth =
@@ -70,6 +88,12 @@ function processColumns<TData>(
         "fixed" in resolvedWidth ? resolvedWidth.fixed : resolvedWidth.weight,
     };
 
+    // First data column is never hideable
+    if (col.kind === "data" && !firstDataColumnSeen) {
+      firstDataColumnSeen = true;
+      clonedDef.enableHiding = false;
+    }
+
     tanstackColumns.push(clonedDef);
 
     const id = col.id;
@@ -113,10 +137,10 @@ function processColumns<TData>(
  *   tc.actions(),
  * ];
  *
- * <DataTable data={items} columns={columns} footer={{ mode: "selection" }} />
+ * <Table data={items} columns={columns} footer={{}} />
  * ```
  */
-export default function DataTable<TData>(props: DataTableProps<TData>) {
+export function Table<TData>(props: DataTableProps<TData>) {
   const {
     data,
     columns,
@@ -126,7 +150,10 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
     initialColumnVisibility,
     draggable,
     footer,
-    size = "regular",
+    size = "lg",
+    variant = "cards",
+    qualifier = "simple",
+    selectionBehavior = "no-select",
     onSelectionChange,
     onRowClick,
     searchTerm,
@@ -138,9 +165,37 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
 
   const effectivePageSize = pageSize ?? (footer ? 10 : data.length);
 
+  // Whether the qualifier column should exist in the DOM.
+  // "simple" only gets a qualifier column for multi-select (checkboxes).
+  // "simple" + no-select/single-select = no qualifier column — single-select
+  // uses row-level background coloring instead.
+  const hasQualifierColumn =
+    qualifier !== "simple" || selectionBehavior === "multi-select";
+
   // 1. Process columns (memoized on columns + size)
   const { tanstackColumns, widthConfig, qualifierColumn, columnKindMap } =
-    useMemo(() => processColumns(columns, size), [columns, size]);
+    useMemo(() => {
+      const processed = processColumns(columns, size);
+      if (!hasQualifierColumn) {
+        // Remove qualifier from TanStack columns and width config entirely
+        return {
+          ...processed,
+          tanstackColumns: processed.tanstackColumns.filter(
+            (c) => c.id !== "qualifier"
+          ),
+          widthConfig: {
+            ...processed.widthConfig,
+            fixedColumnIds: new Set(
+              Array.from(processed.widthConfig.fixedColumnIds).filter(
+                (id) => id !== "qualifier"
+              )
+            ),
+          },
+          qualifierColumn: null,
+        };
+      }
+      return processed;
+    }, [columns, size, hasQualifierColumn]);
 
   // 2. Call useDataTable
   const {
@@ -155,7 +210,9 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
     selectedRowIds,
     clearSelection,
     toggleAllPageRowsSelected,
+    toggleAllRowsSelected,
     isAllPageRowsSelected,
+    isAllRowsSelected,
     isViewingSelected,
     enterViewMode,
     exitViewMode,
@@ -193,16 +250,6 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
       );
     }
   }, [!!serverSide, !!draggable]); // eslint-disable-line react-hooks/exhaustive-deps
-  const footerShowView =
-    footer?.mode === "selection" ? footer.showView : undefined;
-  useEffect(() => {
-    if (process.env.NODE_ENV !== "production" && serverSide && footerShowView) {
-      console.warn(
-        "DataTable: `showView` is ignored when `serverSide` is enabled. " +
-          "View mode requires client-side filtering."
-      );
-    }
-  }, [!!serverSide, !!footerShowView]); // eslint-disable-line react-hooks/exhaustive-deps
   const effectiveDraggable = serverSide ? undefined : draggable;
   const draggableReturn = useDraggableRows({
     data,
@@ -212,10 +259,11 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
   });
 
   const hasDraggable = !!effectiveDraggable;
-  const rowVariant = hasDraggable ? "table" : "list";
 
-  const isSelectable =
-    qualifierColumn != null && qualifierColumn.selectable !== false;
+  const isSelectable = selectionBehavior !== "no-select";
+  const isMultiSelect = selectionBehavior === "multi-select";
+  // Checkboxes appear for any selectable table
+  const showQualifierCheckbox = isSelectable;
 
   // ---------------------------------------------------------------------------
   // Render
@@ -224,11 +272,13 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
   const isServerLoading = !!serverSide?.isLoading;
 
   function renderFooter(footerConfig: DataTableFooterConfig) {
-    if (footerConfig.mode === "selection") {
+    // Mode derived from selectionBehavior — single/multi-select use selection
+    // footer, no-select uses summary footer.
+    if (isSelectable) {
       return (
         <Footer
           mode="selection"
-          multiSelect={footerConfig.multiSelect !== false}
+          multiSelect={isMultiSelect}
           selectionState={selectionState}
           selectedCount={selectedCount}
           onClear={
@@ -239,22 +289,24 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
             })
           }
           onView={
-            footerConfig.showView
+            !serverSide
               ? isViewingSelected
                 ? exitViewMode
                 : enterViewMode
               : undefined
           }
+          isViewingSelected={isViewingSelected}
           pageSize={resolvedPageSize}
           totalItems={totalItems}
           currentPage={currentPage}
           totalPages={totalPages}
           onPageChange={setPage}
+          units={footerConfig.units}
         />
       );
     }
 
-    // Summary mode
+    // Summary mode (no-select only)
     const rangeStart =
       totalItems === 0
         ? 0
@@ -275,6 +327,7 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
         totalPages={totalPages}
         onPageChange={setPage}
         leftExtra={footerConfig.leftExtra}
+        units={footerConfig.units}
       />
     );
   }
@@ -303,7 +356,10 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
               : undefined),
           }}
         >
-          <Table
+          <TableElement
+            size={size}
+            variant={variant}
+            selectionBehavior={selectionBehavior}
             width={
               Object.keys(columnWidths).length > 0
                 ? Object.values(columnWidths).reduce((sum, w) => sum + w, 0)
@@ -311,7 +367,7 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
             }
           >
             <colgroup>
-              {table.getAllLeafColumns().map((col) => (
+              {table.getVisibleLeafColumns().map((col) => (
                 <col
                   key={col.id}
                   style={
@@ -328,28 +384,26 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
                   {headerGroup.headers.map((header, headerIndex) => {
                     const colDef = columnKindMap.get(header.id);
 
-                    // Qualifier header
+                    // Qualifier header — select-all checkbox only for multi-select
                     if (colDef?.kind === "qualifier") {
-                      if (qualifierColumn?.header === false) {
-                        return (
-                          <QualifierContainer key={header.id} type="head" />
-                        );
-                      }
                       return (
                         <QualifierContainer key={header.id} type="head">
-                          <TableQualifier
-                            content={
-                              qualifierColumn?.headerContentType ?? "simple"
-                            }
-                            selectable={isSelectable}
-                            selected={isSelectable && isAllPageRowsSelected}
-                            onSelectChange={
-                              isSelectable
-                                ? (checked) =>
-                                    toggleAllPageRowsSelected(checked)
-                                : undefined
-                            }
-                          />
+                          {isMultiSelect && (
+                            <Checkbox
+                              checked={isAllRowsSelected}
+                              indeterminate={
+                                !isAllRowsSelected && selectedCount > 0
+                              }
+                              onCheckedChange={(checked) => {
+                                // Indeterminate → clear all; otherwise toggle normally
+                                if (!isAllRowsSelected && selectedCount > 0) {
+                                  toggleAllRowsSelected(false);
+                                } else {
+                                  toggleAllRowsSelected(checked);
+                                }
+                              }}
+                            />
+                          )}
                         </QualifierContainer>
                       );
                     }
@@ -437,7 +491,6 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
                       return (
                         <DragOverlayRow
                           row={row}
-                          variant={rowVariant}
                           columnWidths={columnWidths}
                           columnKindMap={columnKindMap}
                           qualifierColumn={qualifierColumn}
@@ -461,7 +514,6 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
                 return (
                   <TableRow
                     key={row.id}
-                    variant={rowVariant}
                     sortableId={rowId}
                     selected={row.getIsSelected()}
                     onClick={() => {
@@ -474,6 +526,10 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
                       if (onRowClick) {
                         onRowClick(row.original);
                       } else if (isSelectable) {
+                        if (!isMultiSelect) {
+                          // single-select: clear all, then select this row
+                          table.toggleAllRowsSelected(false);
+                        }
                         row.toggleSelected();
                       }
                     }}
@@ -484,6 +540,13 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
                       // Qualifier cell
                       if (cellColDef?.kind === "qualifier") {
                         const qDef = cellColDef as OnyxQualifierColumn<TData>;
+
+                        // Resolve content based on the qualifier prop:
+                        // - "simple" renders nothing (checkbox only when selectable)
+                        // - "avatar"/"icon" render from column config
+                        const qualifierContent =
+                          qualifier === "simple" ? "simple" : qDef.content;
+
                         return (
                           <QualifierContainer
                             key={cell.id}
@@ -491,15 +554,20 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
                             onClick={(e) => e.stopPropagation()}
                           >
                             <TableQualifier
-                              content={qDef.content}
+                              content={qualifierContent}
                               initials={qDef.getInitials?.(row.original)}
                               icon={qDef.getIcon?.(row.original)}
                               imageSrc={qDef.getImageSrc?.(row.original)}
-                              selectable={isSelectable}
-                              selected={isSelectable && row.getIsSelected()}
+                              selectable={showQualifierCheckbox}
+                              selected={
+                                showQualifierCheckbox && row.getIsSelected()
+                              }
                               onSelectChange={
-                                isSelectable
+                                showQualifierCheckbox
                                   ? (checked) => {
+                                      if (!isMultiSelect) {
+                                        table.toggleAllRowsSelected(false);
+                                      }
                                       row.toggleSelected(checked);
                                     }
                                   : undefined
@@ -539,7 +607,7 @@ export default function DataTable<TData>(props: DataTableProps<TData>) {
                 );
               })}
             </TableBody>
-          </Table>
+          </TableElement>
         </div>
 
         {footer && renderFooter(footer)}

web/lib/opal/src/components/table/hooks/useDataTable.ts

@@ -153,6 +153,10 @@ interface UseDataTableReturn<TData extends RowData> {
   clearSelection: () => void;
   /** Select or deselect all rows on the current page. */
   toggleAllPageRowsSelected: (selected: boolean) => void;
+  /** Select or deselect all rows across all pages. */
+  toggleAllRowsSelected: (selected: boolean) => void;
+  /** Whether every row across all pages is selected. */
+  isAllRowsSelected: boolean;
 
   // View-mode (filter to selected rows)
   /** Whether the table is currently filtered to show only selected rows. */
@@ -407,6 +411,16 @@ export default function useDataTable<TData extends RowData>(
     table.toggleAllPageRowsSelected(selected);
   };
 
+  // TODO (@raunakab): In server-side mode, these only operate on the loaded
+  // page data, not all rows across all pages. TanStack can't select rows it
+  // doesn't have. Fixing this requires a server-side callback (e.g.
+  // `onSelectAll`) and a `totalItems`-aware selection model.
+  const toggleAllRowsSelected = (selected: boolean) => {
+    table.toggleAllRowsSelected(selected);
+  };
+
+  const isAllRowsSelected = table.getIsAllRowsSelected();
+
   // ---- view mode (filter to selected rows) --------------------------------
   const isViewingSelected = globalFilter.selectedIds != null;
 
@@ -439,8 +453,10 @@ export default function useDataTable<TData extends RowData>(
     selectedCount,
     selectedRowIds,
     isAllPageRowsSelected,
+    isAllRowsSelected,
     clearSelection,
     toggleAllPageRowsSelected,
+    toggleAllRowsSelected,
     isViewingSelected,
     enterViewMode,
     exitViewMode,

web/lib/opal/src/components/table/styles.css

@@ -0,0 +1,164 @@
+/* Imports shared timing tokens (--interactive-duration, --interactive-easing) */
+@import "@opal/core/interactive/shared.css";
+
+/* ---------------------------------------------------------------------------
+ * Table primitives — data-attribute driven styling
+ * Follows the same pattern as card.css / line-item.css.
+ * ------------------------------------------------------------------------- */
+
+/* ---- TableCell ---- */
+
+.tbl-cell[data-size="lg"] {
+  @apply px-1 py-0.5;
+}
+.tbl-cell[data-size="md"] {
+  @apply pl-0.5 pr-1.5 py-1.5;
+}
+
+.tbl-cell-inner[data-size="lg"] {
+  @apply h-10 px-1;
+}
+.tbl-cell-inner[data-size="md"] {
+  @apply h-6 px-0.5;
+}
+
+/* ---- TableHead ---- */
+
+.table-head {
+  @apply relative sticky top-0 z-20;
+  background: var(--table-header-bg, transparent);
+}
+.table-head[data-size="lg"] {
+  @apply px-2 py-1;
+}
+.table-head[data-size="md"] {
+  @apply px-2 py-1;
+}
+.table-head[data-bottom-border] {
+  @apply border-b border-transparent hover:border-border-03;
+}
+
+/* Inner text wrapper */
+.table-head[data-size="lg"] .table-head-label {
+  @apply py-2 px-0.5;
+}
+.table-head[data-size="md"] .table-head-label {
+  @apply py-1;
+}
+
+/* Sort button wrapper */
+.table-head[data-size="lg"] .table-head-sort {
+  @apply py-1.5;
+}
+
+/* ---- TableRow (base) ---- */
+
+.tbl-row > td {
+  @apply bg-background-tint-00;
+  transition: background-color var(--interactive-duration)
+    var(--interactive-easing);
+}
+
+.tbl-row[data-selected] > td {
+  @apply bg-[var(--action-link-01)];
+}
+
+.tbl-row[data-disabled] {
+  @apply pointer-events-none;
+}
+
+/* Suppress default focus ring on rows — the row bg is the indicator */
+.tbl-row:focus,
+.tbl-row:focus-visible {
+  outline: none;
+}
+
+/* ---- variant="rows" — traditional borders, no gaps ---- */
+
+table[data-variant="rows"] .tbl-row > td {
+  @apply border-b border-border-01;
+}
+
+/* Hover/focus only for selectable tables */
+table[data-variant="rows"][data-selection="single-select"] .tbl-row,
+table[data-variant="rows"][data-selection="multi-select"] .tbl-row {
+  @apply cursor-pointer;
+}
+table[data-variant="rows"][data-selection="single-select"] .tbl-row:hover > td,
+table[data-variant="rows"][data-selection="multi-select"] .tbl-row:hover > td {
+  @apply bg-background-tint-02;
+}
+table[data-variant="rows"] .tbl-row:focus-visible > td,
+table[data-variant="rows"] .tbl-row:has(:focus-visible) > td {
+  @apply bg-action-link-01;
+}
+
+/* ---- variant="cards" — rounded cards with gap ---- */
+
+table[data-variant="cards"] .tbl-row > td {
+  @apply bg-clip-padding border-y-[2px] border-x-0 border-transparent;
+}
+table[data-variant="cards"] .tbl-row > td:first-child {
+  @apply rounded-l-12;
+}
+table[data-variant="cards"] .tbl-row > td:last-child {
+  @apply rounded-r-12;
+}
+
+/* When a drag handle is present the second-to-last td gets the rounding */
+table[data-variant="cards"] .tbl-row[data-drag-handle] > td:nth-last-child(2) {
+  @apply rounded-r-12;
+}
+table[data-variant="cards"] .tbl-row[data-drag-handle] > td:last-child {
+  border-radius: 0;
+}
+
+/* Hover/focus only for selectable tables */
+table[data-variant="cards"][data-selection="single-select"] .tbl-row,
+table[data-variant="cards"][data-selection="multi-select"] .tbl-row {
+  @apply cursor-pointer;
+}
+table[data-variant="cards"][data-selection="single-select"] .tbl-row:hover > td,
+table[data-variant="cards"][data-selection="multi-select"] .tbl-row:hover > td {
+  @apply bg-background-tint-02;
+}
+table[data-variant="cards"] .tbl-row:focus-visible > td,
+table[data-variant="cards"] .tbl-row:has(:focus-visible) > td {
+  @apply bg-action-link-01;
+}
+
+/* ---- QualifierContainer ---- */
+
+.tbl-qualifier[data-type="head"] {
+  @apply w-px whitespace-nowrap py-1 sticky top-0 z-20;
+  background: var(--table-header-bg, transparent);
+}
+.tbl-qualifier[data-type="head"][data-size="md"] {
+  @apply py-0.5;
+}
+
+.tbl-qualifier[data-type="cell"] {
+  @apply w-px whitespace-nowrap py-1;
+}
+.tbl-qualifier[data-type="cell"][data-size="md"] {
+  @apply py-0.5;
+}
+
+/* ---- ActionsContainer ---- */
+
+.tbl-actions {
+  @apply sticky right-0 w-px whitespace-nowrap px-1;
+}
+.tbl-actions[data-type="head"] {
+  @apply z-30 sticky top-0 px-2 py-1;
+  background: var(--table-header-bg, transparent);
+}
+
+/* ---- Footer ---- */
+
+.table-footer[data-size="lg"] {
+  @apply min-h-[2.75rem];
+}
+.table-footer[data-size="md"] {
+  @apply min-h-[2.25rem];
+}

web/lib/opal/src/components/table/types.ts

@@ -4,9 +4,10 @@ import type {
   SortingState,
   VisibilityState,
 } from "@tanstack/react-table";
-import type { TableSize } from "@/refresh-components/table/TableSizeContext";
+import type { TableSize } from "@opal/components/table/TableSizeContext";
+import type { TableVariant } from "@opal/components/table/TableElement";
 import type { IconFunctionComponent } from "@opal/types";
-import type { SortDirection } from "@/refresh-components/table/TableHead";
+import type { SortDirection } from "@opal/components/table/TableHead";
 
 // ---------------------------------------------------------------------------
 // Column width (mirrors useColumnWidths types)
@@ -129,26 +130,16 @@ export interface DataTableDraggableConfig {
   ) => void | Promise<void>;
 }
 
-export interface DataTableFooterSelection {
-  mode: "selection";
-  /** Whether the table supports selecting multiple rows. @default true */
-  multiSelect?: boolean;
-  /** When true, shows a "View" button that filters the table to only selected rows. @default false */
-  showView?: boolean;
-  /** Handler for the "Clear" button. When omitted, the default clearSelection is used. */
+/** Footer configuration. Mode is derived from `selectionBehavior` automatically. */
+export interface DataTableFooterConfig {
+  /** Handler for the "Clear" button (multi-select only). When omitted, the default clearSelection is used. */
   onClear?: () => void;
-}
-
-export interface DataTableFooterSummary {
-  mode: "summary";
-  /** Optional extra element rendered after the summary text (e.g. a download icon). */
+  /** Unit label for count pagination, e.g. "users", "documents" (multi-select only). */
+  units?: string;
+  /** Optional extra element rendered after the summary text, e.g. a download icon (summary mode only). */
   leftExtra?: ReactNode;
 }
 
-export type DataTableFooterConfig =
-  | DataTableFooterSelection
-  | DataTableFooterSummary;
-
 export interface DataTableProps<TData> {
   /** Row data array. */
   data: TData[];
@@ -166,8 +157,10 @@ export interface DataTableProps<TData> {
   draggable?: DataTableDraggableConfig;
   /** Footer configuration. */
   footer?: DataTableFooterConfig;
-  /** Table size variant. @default "regular" */
+  /** Table size variant. @default "lg" */
   size?: TableSize;
+  /** Visual row variant. @default "cards" */
+  variant?: TableVariant;
   /** Called whenever the set of selected row IDs changes. Receives IDs produced by `getRowId`. */
   onSelectionChange?: (selectedIds: string[]) => void;
   /** Called when a row is clicked (replaces the default selection toggle). */

web/lib/opal/src/core/animations/Hoverable.stories.tsx

@@ -0,0 +1,162 @@
+import type { Meta, StoryObj } from "@storybook/react";
+import { Hoverable } from "@opal/core";
+
+// ---------------------------------------------------------------------------
+// Meta
+// ---------------------------------------------------------------------------
+
+const meta: Meta = {
+  title: "Core/Hoverable",
+  tags: ["autodocs"],
+  parameters: {
+    layout: "centered",
+  },
+};
+
+export default meta;
+
+// ---------------------------------------------------------------------------
+// Stories
+// ---------------------------------------------------------------------------
+
+/** Group mode — hovering the root reveals hidden items. */
+export const GroupMode: StoryObj = {
+  render: () => (
+    <Hoverable.Root group="demo">
+      <div
+        style={{
+          display: "flex",
+          alignItems: "center",
+          gap: "0.75rem",
+          padding: "1rem",
+          border: "1px solid var(--border-02)",
+          borderRadius: "0.5rem",
+          minWidth: 260,
+        }}
+      >
+        <span style={{ color: "var(--text-01)" }}>Hover this card</span>
+        <Hoverable.Item group="demo" variant="opacity-on-hover">
+          <span style={{ color: "var(--text-03)" }}>✓ Revealed</span>
+        </Hoverable.Item>
+      </div>
+    </Hoverable.Root>
+  ),
+};
+
+/** Local mode — hovering the item itself reveals it (no Root needed). */
+export const LocalMode: StoryObj = {
+  render: () => (
+    <div
+      style={{
+        display: "flex",
+        alignItems: "center",
+        gap: "0.75rem",
+        padding: "1rem",
+      }}
+    >
+      <span style={{ color: "var(--text-01)" }}>Hover the icon →</span>
+      <Hoverable.Item variant="opacity-on-hover">
+        <span style={{ fontSize: "1.25rem" }}>🗑</span>
+      </Hoverable.Item>
+    </div>
+  ),
+};
+
+/** Multiple independent groups on the same page. */
+export const MultipleGroups: StoryObj = {
+  render: () => (
+    <div style={{ display: "flex", flexDirection: "column", gap: "0.75rem" }}>
+      {(["alpha", "beta"] as const).map((group) => (
+        <Hoverable.Root key={group} group={group}>
+          <div
+            style={{
+              display: "flex",
+              alignItems: "center",
+              gap: "0.75rem",
+              padding: "1rem",
+              border: "1px solid var(--border-02)",
+              borderRadius: "0.5rem",
+            }}
+          >
+            <span style={{ color: "var(--text-01)" }}>Group: {group}</span>
+            <Hoverable.Item group={group} variant="opacity-on-hover">
+              <span style={{ color: "var(--text-03)" }}>✓ Revealed</span>
+            </Hoverable.Item>
+          </div>
+        </Hoverable.Root>
+      ))}
+    </div>
+  ),
+};
+
+/** Multiple items revealed by a single root. */
+export const MultipleItems: StoryObj = {
+  render: () => (
+    <Hoverable.Root group="multi">
+      <div
+        style={{
+          display: "flex",
+          alignItems: "center",
+          gap: "0.75rem",
+          padding: "1rem",
+          border: "1px solid var(--border-02)",
+          borderRadius: "0.5rem",
+        }}
+      >
+        <span style={{ color: "var(--text-01)" }}>Hover to reveal all</span>
+        <Hoverable.Item group="multi" variant="opacity-on-hover">
+          <span>Edit</span>
+        </Hoverable.Item>
+        <Hoverable.Item group="multi" variant="opacity-on-hover">
+          <span>Delete</span>
+        </Hoverable.Item>
+        <Hoverable.Item group="multi" variant="opacity-on-hover">
+          <span>Share</span>
+        </Hoverable.Item>
+      </div>
+    </Hoverable.Root>
+  ),
+};
+
+/** Nested groups — inner and outer hover independently. */
+export const NestedGroups: StoryObj = {
+  render: () => (
+    <Hoverable.Root group="outer">
+      <div
+        style={{
+          padding: "1rem",
+          border: "1px solid var(--border-02)",
+          borderRadius: "0.5rem",
+          display: "flex",
+          flexDirection: "column",
+          gap: "0.75rem",
+        }}
+      >
+        <div style={{ display: "flex", alignItems: "center", gap: "0.75rem" }}>
+          <span style={{ color: "var(--text-01)" }}>Outer card</span>
+          <Hoverable.Item group="outer" variant="opacity-on-hover">
+            <span style={{ color: "var(--text-03)" }}>Outer action</span>
+          </Hoverable.Item>
+        </div>
+
+        <Hoverable.Root group="inner">
+          <div
+            style={{
+              display: "flex",
+              alignItems: "center",
+              gap: "0.75rem",
+              padding: "0.75rem",
+              border: "1px solid var(--border-03)",
+              borderRadius: "0.375rem",
+            }}
+          >
+            <span style={{ color: "var(--text-02)" }}>Inner card</span>
+            <Hoverable.Item group="inner" variant="opacity-on-hover">
+              <span style={{ color: "var(--text-03)" }}>Inner action</span>
+            </Hoverable.Item>
+          </div>
+        </Hoverable.Root>
+      </div>
+    </Hoverable.Root>
+  ),
+};

web/lib/opal/src/core/hoverable/Hoverable.stories.tsx

@@ -1,127 +0,0 @@
-import type { Meta, StoryObj } from "@storybook/react";
-import { Hoverable } from "@opal/core";
-import SvgX from "@opal/icons/x";
-
-// ---------------------------------------------------------------------------
-// Shared styles
-// ---------------------------------------------------------------------------
-
-const cardStyle: React.CSSProperties = {
-  display: "flex",
-  alignItems: "center",
-  justifyContent: "space-between",
-  gap: "1rem",
-  padding: "0.75rem 1rem",
-  borderRadius: "0.5rem",
-  border: "1px solid var(--border-02)",
-  background: "var(--background-neutral-01)",
-  minWidth: 220,
-};
-
-const labelStyle: React.CSSProperties = {
-  fontSize: "0.875rem",
-  fontWeight: 500,
-};
-
-// ---------------------------------------------------------------------------
-// Meta
-// ---------------------------------------------------------------------------
-
-const meta: Meta = {
-  title: "Core/Hoverable",
-  tags: ["autodocs"],
-  parameters: {
-    layout: "centered",
-  },
-};
-
-export default meta;
-
-// ---------------------------------------------------------------------------
-// Stories
-// ---------------------------------------------------------------------------
-
-/**
- * Local hover mode -- no `group` prop on the Item.
- * The icon only appears when you hover directly over the Item element itself.
- */
-export const LocalHover: StoryObj = {
-  render: () => (
-    <div style={cardStyle}>
-      <span style={labelStyle}>Hover this card area</span>
-
-      <Hoverable.Item variant="opacity-on-hover">
-        <SvgX width={16} height={16} />
-      </Hoverable.Item>
-    </div>
-  ),
-};
-
-/**
- * Group hover mode -- hovering anywhere inside the Root reveals the Item.
- */
-export const GroupHover: StoryObj = {
-  render: () => (
-    <Hoverable.Root group="card">
-      <div style={cardStyle}>
-        <span style={labelStyle}>Hover anywhere on this card</span>
-
-        <Hoverable.Item group="card" variant="opacity-on-hover">
-          <SvgX width={16} height={16} />
-        </Hoverable.Item>
-      </div>
-    </Hoverable.Root>
-  ),
-};
-
-/**
- * Nested groups demonstrating isolation.
- *
- * - Hovering the outer card reveals only the outer icon.
- * - Hovering the inner card reveals only the inner icon.
- */
-export const NestedGroups: StoryObj = {
-  render: () => (
-    <Hoverable.Root group="outer">
-      <div
-        style={{
-          ...cardStyle,
-          flexDirection: "column",
-          alignItems: "stretch",
-          gap: "0.75rem",
-          padding: "1rem",
-          minWidth: 300,
-        }}
-      >
-        <div
-          style={{
-            display: "flex",
-            alignItems: "center",
-            justifyContent: "space-between",
-          }}
-        >
-          <span style={labelStyle}>Outer card</span>
-
-          <Hoverable.Item group="outer" variant="opacity-on-hover">
-            <SvgX width={16} height={16} />
-          </Hoverable.Item>
-        </div>
-
-        <Hoverable.Root group="inner">
-          <div
-            style={{
-              ...cardStyle,
-              background: "var(--background-neutral-02)",
-            }}
-          >
-            <span style={labelStyle}>Inner card</span>
-
-            <Hoverable.Item group="inner" variant="opacity-on-hover">
-              <SvgX width={16} height={16} />
-            </Hoverable.Item>
-          </div>
-        </Hoverable.Root>
-      </div>
-    </Hoverable.Root>
-  ),
-};

web/lib/opal/src/core/interactive/shared.css

@@ -15,13 +15,21 @@
   initial-value: transparent;
 }
 
+/* Shared timing tokens — used by .interactive and other surfaces (e.g. table rows) */
+:root {
+  --interactive-duration: 150ms;
+  --interactive-easing: ease-in-out;
+}
+
 /* Base interactive surface */
 .interactive {
   @apply cursor-pointer select-none;
   transition:
-    background-color 150ms ease-in-out,
-    --interactive-foreground 150ms ease-in-out,
-    --interactive-foreground-icon 150ms ease-in-out;
+    background-color var(--interactive-duration) var(--interactive-easing),
+    --interactive-foreground var(--interactive-duration)
+      var(--interactive-easing),
+    --interactive-foreground-icon var(--interactive-duration)
+      var(--interactive-easing);
 }
 .interactive[data-disabled] {
   @apply cursor-not-allowed;