fix: register OnyxError exception handler in model_server

Without this, any OnyxError raised in model_server endpoints would
result in an unhandled 500 instead of the proper error response.

.github/workflows/deployment.yml

@@ -151,7 +151,7 @@ jobs:
           fetch-depth: 0
 
       - name: Setup uv
-        uses: astral-sh/setup-uv@61cb8a9741eeb8a550a1b8544337180c0fc8476b # ratchet:astral-sh/setup-uv@v7
+        uses: astral-sh/setup-uv@5a095e7a2014a4212f075830d4f7277575a9d098 # ratchet:astral-sh/setup-uv@v7
         with:
           version: "0.9.9"
           # NOTE: This isn't caching much and zizmor suggests this could be poisoned, so disable.

.github/workflows/post-merge-beta-cherry-pick.yml

@@ -70,7 +70,7 @@ jobs:
 
       - name: Install the latest version of uv
         if: steps.gate.outputs.should_cherrypick == 'true'
-        uses: astral-sh/setup-uv@61cb8a9741eeb8a550a1b8544337180c0fc8476b # ratchet:astral-sh/setup-uv@v7
+        uses: astral-sh/setup-uv@5a095e7a2014a4212f075830d4f7277575a9d098 # ratchet:astral-sh/setup-uv@v7
         with:
           enable-cache: false
           version: "0.9.9"

.github/workflows/pr-playwright-tests.yml

@@ -468,7 +468,7 @@ jobs:
 
       - name: Install the latest version of uv
         if: always()
-        uses: astral-sh/setup-uv@61cb8a9741eeb8a550a1b8544337180c0fc8476b # ratchet:astral-sh/setup-uv@v7
+        uses: astral-sh/setup-uv@5a095e7a2014a4212f075830d4f7277575a9d098 # ratchet:astral-sh/setup-uv@v7
         with:
           enable-cache: false
           version: "0.9.9"
@@ -707,7 +707,7 @@ jobs:
       pull-requests: write
     steps:
       - name: Download visual diff summaries
-        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131
+        uses: actions/download-artifact@70fc10c6e5e1ce46ad2ea6f2b72d43f7d47b13c3
         with:
           pattern: screenshot-diff-summary-*
           path: summaries/

.github/workflows/pr-quality-checks.yml

@@ -28,7 +28,7 @@ jobs:
         with:
           python-version: "3.11"
       - name: Setup Terraform
-        uses: hashicorp/setup-terraform@b9cd54a3c349d3f38e8881555d616ced269862dd # ratchet:hashicorp/setup-terraform@v3
+        uses: hashicorp/setup-terraform@5e8dbf3c6d9deaf4193ca7a8fb23f2ac83bb6c85 # ratchet:hashicorp/setup-terraform@v4.0.0
       - name: Setup node
         uses: actions/setup-node@6044e13b5dc448c55e2357c09f80417699197238 # ratchet:actions/setup-node@v6
         with: # zizmor: ignore[cache-poisoning]

.github/workflows/release-devtools.yml

@@ -26,7 +26,7 @@ jobs:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # ratchet:actions/checkout@v6
         with:
           persist-credentials: false
-      - uses: astral-sh/setup-uv@61cb8a9741eeb8a550a1b8544337180c0fc8476b # ratchet:astral-sh/setup-uv@v7
+      - uses: astral-sh/setup-uv@5a095e7a2014a4212f075830d4f7277575a9d098 # ratchet:astral-sh/setup-uv@v7
         with:
           enable-cache: false
           version: "0.9.9"

.github/workflows/zizmor.yml

@@ -24,7 +24,7 @@ jobs:
           persist-credentials: false
 
       - name: Install the latest version of uv
-        uses: astral-sh/setup-uv@61cb8a9741eeb8a550a1b8544337180c0fc8476b # ratchet:astral-sh/setup-uv@v7
+        uses: astral-sh/setup-uv@5a095e7a2014a4212f075830d4f7277575a9d098 # ratchet:astral-sh/setup-uv@v7
         with:
           enable-cache: false
           version: "0.9.9"

backend/Dockerfile

@@ -141,6 +141,7 @@ COPY --chown=onyx:onyx ./scripts/debugging /app/scripts/debugging
 COPY --chown=onyx:onyx ./scripts/force_delete_connector_by_id.py /app/scripts/force_delete_connector_by_id.py
 COPY --chown=onyx:onyx ./scripts/supervisord_entrypoint.sh /app/scripts/supervisord_entrypoint.sh
 COPY --chown=onyx:onyx ./scripts/setup_craft_templates.sh /app/scripts/setup_craft_templates.sh
+COPY --chown=onyx:onyx ./scripts/reencrypt_secrets.py /app/scripts/reencrypt_secrets.py
 RUN chmod +x /app/scripts/supervisord_entrypoint.sh /app/scripts/setup_craft_templates.sh
 
 # Run Craft template setup at build time when ENABLE_CRAFT=true

backend/alembic_tenants/versions/3b9f09038764_add_read_only_kg_user.py

@@ -11,7 +11,6 @@ from sqlalchemy import text
 from alembic import op
 from onyx.configs.app_configs import DB_READONLY_PASSWORD
 from onyx.configs.app_configs import DB_READONLY_USER
-from shared_configs.configs import MULTI_TENANT
 
 
 # revision identifiers, used by Alembic.
@@ -22,59 +21,52 @@ depends_on = None
 
 
 def upgrade() -> None:
-    if MULTI_TENANT:
+    # Enable pg_trgm extension if not already enabled
+    op.execute("CREATE EXTENSION IF NOT EXISTS pg_trgm")
 
-        # Enable pg_trgm extension if not already enabled
-        op.execute("CREATE EXTENSION IF NOT EXISTS pg_trgm")
+    # Create the read-only db user if it does not already exist.
+    if not (DB_READONLY_USER and DB_READONLY_PASSWORD):
+        raise Exception("DB_READONLY_USER or DB_READONLY_PASSWORD is not set")
 
-        # Create read-only db user here only in multi-tenant mode. For single-tenant mode,
-        # the user is created in the standard migration.
-        if not (DB_READONLY_USER and DB_READONLY_PASSWORD):
-            raise Exception("DB_READONLY_USER or DB_READONLY_PASSWORD is not set")
-
-        op.execute(
-            text(
-                f"""
-                DO $$
-                BEGIN
-                    -- Check if the read-only user already exists
-                    IF NOT EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname = '{DB_READONLY_USER}') THEN
-                        -- Create the read-only user with the specified password
-                        EXECUTE format('CREATE USER %I WITH PASSWORD %L', '{DB_READONLY_USER}', '{DB_READONLY_PASSWORD}');
-                        -- First revoke all privileges to ensure a clean slate
-                        EXECUTE format('REVOKE ALL ON DATABASE %I FROM %I', current_database(), '{DB_READONLY_USER}');
-                        -- Grant only the CONNECT privilege to allow the user to connect to the database
-                        -- but not perform any operations without additional specific grants
-                        EXECUTE format('GRANT CONNECT ON DATABASE %I TO %I', current_database(), '{DB_READONLY_USER}');
-                    END IF;
-                END
-                $$;
-                """
-            )
-        )
-
-
-def downgrade() -> None:
-    if MULTI_TENANT:
-        # Drop read-only db user here only in single tenant mode. For multi-tenant mode,
-        # the user is dropped in the alembic_tenants migration.
-
-        op.execute(
-            text(
-                f"""
+    op.execute(
+        text(
+            f"""
             DO $$
             BEGIN
-                IF EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname = '{DB_READONLY_USER}') THEN
-                    -- First revoke all privileges from the database
+                -- Check if the read-only user already exists
+                IF NOT EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname = '{DB_READONLY_USER}') THEN
+                    -- Create the read-only user with the specified password
+                    EXECUTE format('CREATE USER %I WITH PASSWORD %L', '{DB_READONLY_USER}', '{DB_READONLY_PASSWORD}');
+                    -- First revoke all privileges to ensure a clean slate
                     EXECUTE format('REVOKE ALL ON DATABASE %I FROM %I', current_database(), '{DB_READONLY_USER}');
-                    -- Then revoke all privileges from the public schema
-                    EXECUTE format('REVOKE ALL ON SCHEMA public FROM %I', '{DB_READONLY_USER}');
-                    -- Then drop the user
-                    EXECUTE format('DROP USER %I', '{DB_READONLY_USER}');
+                    -- Grant only the CONNECT privilege to allow the user to connect to the database
+                    -- but not perform any operations without additional specific grants
+                    EXECUTE format('GRANT CONNECT ON DATABASE %I TO %I', current_database(), '{DB_READONLY_USER}');
                 END IF;
             END
             $$;
-        """
-            )
+            """
         )
-        op.execute(text("DROP EXTENSION IF EXISTS pg_trgm"))
+    )
+
+
+def downgrade() -> None:
+    op.execute(
+        text(
+            f"""
+        DO $$
+        BEGIN
+            IF EXISTS (SELECT FROM pg_catalog.pg_roles WHERE rolname = '{DB_READONLY_USER}') THEN
+                -- First revoke all privileges from the database
+                EXECUTE format('REVOKE ALL ON DATABASE %I FROM %I', current_database(), '{DB_READONLY_USER}');
+                -- Then revoke all privileges from the public schema
+                EXECUTE format('REVOKE ALL ON SCHEMA public FROM %I', '{DB_READONLY_USER}');
+                -- Then drop the user
+                EXECUTE format('DROP USER %I', '{DB_READONLY_USER}');
+            END IF;
+        END
+        $$;
+    """
+        )
+    )
+    op.execute(text("DROP EXTENSION IF EXISTS pg_trgm"))

backend/ee/onyx/external_permissions/google_drive/doc_sync.py

@@ -68,6 +68,7 @@ def get_external_access_for_raw_gdrive_file(
     company_domain: str,
     retriever_drive_service: GoogleDriveService | None,
     admin_drive_service: GoogleDriveService,
+    fallback_user_email: str,
     add_prefix: bool = False,
 ) -> ExternalAccess:
     """
@@ -79,6 +80,11 @@ def get_external_access_for_raw_gdrive_file(
                 set add_prefix to True so group IDs are prefixed with the source type.
                 When invoked from doc_sync (permission sync), use the default (False)
                 since upsert_document_external_perms handles prefixing.
+    fallback_user_email: When we cannot retrieve any permission info for a file
+                (e.g. externally-owned files where the API returns no permissions
+                and permissions.list returns 403), fall back to granting access
+                to this user. This is typically the impersonated org user whose
+                drive contained the file.
     """
     doc_id = file.get("id")
     if not doc_id:
@@ -117,6 +123,26 @@ def get_external_access_for_raw_gdrive_file(
                 [permissions_list, backup_permissions_list]
             )
 
+    # For externally-owned files, the Drive API may return no permissions
+    # and permissions.list may return 403. In this case, fall back to
+    # granting access to the user who found the file in their drive.
+    # Note, even if other users also have access to this file,
+    # they will not be granted access in Onyx.
+    # We check permissions_list (the final result after all fetch attempts)
+    # rather than the raw fields, because permission_ids may be present
+    # but the actual fetch can still return empty due to a 403.
+    if not permissions_list:
+        logger.info(
+            f"No permission info available for file {doc_id} "
+            f"(likely owned by a user outside of your organization). "
+            f"Falling back to granting access to retriever user: {fallback_user_email}"
+        )
+        return ExternalAccess(
+            external_user_emails={fallback_user_email},
+            external_user_group_ids=set(),
+            is_public=False,
+        )
+
     folder_ids_to_inherit_permissions_from: set[str] = set()
     user_emails: set[str] = set()
     group_emails: set[str] = set()

backend/ee/onyx/utils/encryption.py

@@ -14,67 +14,91 @@ from onyx.utils.variable_functionality import fetch_versioned_implementation
 logger = setup_logger()
 
 
-@lru_cache(maxsize=1)
+@lru_cache(maxsize=2)
 def _get_trimmed_key(key: str) -> bytes:
     encoded_key = key.encode()
     key_length = len(encoded_key)
     if key_length < 16:
         raise RuntimeError("Invalid ENCRYPTION_KEY_SECRET - too short")
-    elif key_length > 32:
-        key = key[:32]
-    elif key_length not in (16, 24, 32):
-        valid_lengths = [16, 24, 32]
-        key = key[: min(valid_lengths, key=lambda x: abs(x - key_length))]
 
-    return encoded_key
+    # Trim to the largest valid AES key size that fits
+    valid_lengths = [32, 24, 16]
+    for size in valid_lengths:
+        if key_length >= size:
+            return encoded_key[:size]
+
+    raise AssertionError("unreachable")
 
 
-def _encrypt_string(input_str: str) -> bytes:
-    if not ENCRYPTION_KEY_SECRET:
+def _encrypt_string(input_str: str, key: str | None = None) -> bytes:
+    effective_key = key if key is not None else ENCRYPTION_KEY_SECRET
+    if not effective_key:
         return input_str.encode()
 
-    key = _get_trimmed_key(ENCRYPTION_KEY_SECRET)
+    trimmed = _get_trimmed_key(effective_key)
     iv = urandom(16)
     padder = padding.PKCS7(algorithms.AES.block_size).padder()
     padded_data = padder.update(input_str.encode()) + padder.finalize()
 
-    cipher = Cipher(algorithms.AES(key), modes.CBC(iv), backend=default_backend())
+    cipher = Cipher(algorithms.AES(trimmed), modes.CBC(iv), backend=default_backend())
     encryptor = cipher.encryptor()
     encrypted_data = encryptor.update(padded_data) + encryptor.finalize()
 
     return iv + encrypted_data
 
 
-def _decrypt_bytes(input_bytes: bytes) -> str:
-    if not ENCRYPTION_KEY_SECRET:
+def _decrypt_bytes(input_bytes: bytes, key: str | None = None) -> str:
+    effective_key = key if key is not None else ENCRYPTION_KEY_SECRET
+    if not effective_key:
         return input_bytes.decode()
 
-    key = _get_trimmed_key(ENCRYPTION_KEY_SECRET)
-    iv = input_bytes[:16]
-    encrypted_data = input_bytes[16:]
+    trimmed = _get_trimmed_key(effective_key)
+    try:
+        iv = input_bytes[:16]
+        encrypted_data = input_bytes[16:]
 
-    cipher = Cipher(algorithms.AES(key), modes.CBC(iv), backend=default_backend())
-    decryptor = cipher.decryptor()
-    decrypted_padded_data = decryptor.update(encrypted_data) + decryptor.finalize()
+        cipher = Cipher(
+            algorithms.AES(trimmed), modes.CBC(iv), backend=default_backend()
+        )
+        decryptor = cipher.decryptor()
+        decrypted_padded_data = decryptor.update(encrypted_data) + decryptor.finalize()
 
-    unpadder = padding.PKCS7(algorithms.AES.block_size).unpadder()
-    decrypted_data = unpadder.update(decrypted_padded_data) + unpadder.finalize()
+        unpadder = padding.PKCS7(algorithms.AES.block_size).unpadder()
+        decrypted_data = unpadder.update(decrypted_padded_data) + unpadder.finalize()
 
-    return decrypted_data.decode()
+        return decrypted_data.decode()
+    except (ValueError, UnicodeDecodeError):
+        if key is not None:
+            # Explicit key was provided — don't fall back silently
+            raise
+        # Read path: attempt raw UTF-8 decode as a fallback for legacy data.
+        # Does NOT handle data encrypted with a different key — that
+        # ciphertext is not valid UTF-8 and will raise below.
+        logger.warning(
+            "AES decryption failed — falling back to raw decode. "
+            "Run the re-encrypt secrets script to rotate to the current key."
+        )
+        try:
+            return input_bytes.decode()
+        except UnicodeDecodeError:
+            raise ValueError(
+                "Data is not valid UTF-8 — likely encrypted with a different key. "
+                "Run the re-encrypt secrets script to rotate to the current key."
+            ) from None
 
 
-def encrypt_string_to_bytes(input_str: str) -> bytes:
+def encrypt_string_to_bytes(input_str: str, key: str | None = None) -> bytes:
     versioned_encryption_fn = fetch_versioned_implementation(
         "onyx.utils.encryption", "_encrypt_string"
     )
-    return versioned_encryption_fn(input_str)
+    return versioned_encryption_fn(input_str, key=key)
 
 
-def decrypt_bytes_to_string(input_bytes: bytes) -> str:
+def decrypt_bytes_to_string(input_bytes: bytes, key: str | None = None) -> str:
     versioned_decryption_fn = fetch_versioned_implementation(
         "onyx.utils.encryption", "_decrypt_bytes"
     )
-    return versioned_decryption_fn(input_bytes)
+    return versioned_decryption_fn(input_bytes, key=key)
 
 
 def test_encryption() -> None:

backend/model_server/encoders.py

@@ -4,10 +4,11 @@ from typing import Any
 from typing import TYPE_CHECKING
 
 from fastapi import APIRouter
-from fastapi import HTTPException
 from fastapi import Request
 
 from model_server.utils import simple_log_function_time
+from onyx.error_handling.error_codes import OnyxErrorCode
+from onyx.error_handling.exceptions import OnyxError
 from onyx.utils.logger import setup_logger
 from shared_configs.enums import EmbedTextType
 from shared_configs.model_server_models import Embedding
@@ -188,7 +189,7 @@ async def process_embed_request(
         )
 
     if not embed_request.texts:
-        raise HTTPException(status_code=400, detail="No texts to be embedded")
+        raise OnyxError(OnyxErrorCode.VALIDATION_ERROR, "No texts to be embedded")
 
     if not all(embed_request.texts):
         raise ValueError("Empty strings are not allowed for embedding.")
@@ -211,14 +212,12 @@ async def process_embed_request(
         )
         return EmbedResponse(embeddings=embeddings)
     except RateLimitError as e:
-        raise HTTPException(
-            status_code=429,
-            detail=str(e),
-        )
+        raise OnyxError(OnyxErrorCode.RATE_LIMITED, str(e))
     except Exception as e:
         logger.exception(
             f"Error during embedding process: provider={embed_request.provider_type} model={embed_request.model_name}"
         )
-        raise HTTPException(
-            status_code=500, detail=f"Error during embedding process: {e}"
+        raise OnyxError(
+            OnyxErrorCode.INTERNAL_ERROR,
+            f"Error during embedding process: {e}",
         )

backend/model_server/main.py

@@ -18,6 +18,7 @@ from model_server.encoders import router as encoders_router
 from model_server.management_endpoints import router as management_router
 from model_server.utils import get_gpu_type
 from onyx import __version__
+from onyx.error_handling.exceptions import register_onyx_exception_handlers
 from onyx.utils.logger import setup_logger
 from onyx.utils.logger import setup_uvicorn_logger
 from onyx.utils.middleware import add_onyx_request_id_middleware
@@ -108,6 +109,8 @@ def get_model_app() -> FastAPI:
     application.include_router(management_router)
     application.include_router(encoders_router)
 
+    register_onyx_exception_handlers(application)
+
     request_id_prefix = "INF"
     if INDEXING_ONLY:
         request_id_prefix = "IDX"

backend/onyx/chat/llm_step.py

@@ -15,6 +15,7 @@ from onyx.chat.citation_processor import DynamicCitationProcessor
 from onyx.chat.emitter import Emitter
 from onyx.chat.models import ChatMessageSimple
 from onyx.chat.models import LlmStepResult
+from onyx.chat.tool_call_args_streaming import maybe_emit_argument_delta
 from onyx.configs.app_configs import LOG_ONYX_MODEL_INTERACTIONS
 from onyx.configs.app_configs import PROMPT_CACHE_CHAT_HISTORY
 from onyx.configs.constants import MessageType
@@ -54,6 +55,7 @@ from onyx.server.query_and_chat.streaming_models import ReasoningStart
 from onyx.tools.models import ToolCallKickoff
 from onyx.tracing.framework.create import generation_span
 from onyx.utils.b64 import get_image_type_from_bytes
+from onyx.utils.jsonriver import Parser
 from onyx.utils.logger import setup_logger
 from onyx.utils.postgres_sanitization import sanitize_string
 from onyx.utils.text_processing import find_all_json_objects
@@ -1009,6 +1011,7 @@ def run_llm_step_pkt_generator(
         )
 
     id_to_tool_call_map: dict[int, dict[str, Any]] = {}
+    arg_parsers: dict[int, Parser] = {}
     reasoning_start = False
     answer_start = False
     accumulated_reasoning = ""
@@ -1215,7 +1218,14 @@ def run_llm_step_pkt_generator(
                 yield from _close_reasoning_if_active()
 
                 for tool_call_delta in delta.tool_calls:
+                    # maybe_emit depends and update being called first and attaching the delta
                     _update_tool_call_with_delta(id_to_tool_call_map, tool_call_delta)
+                    yield from maybe_emit_argument_delta(
+                        tool_calls_in_progress=id_to_tool_call_map,
+                        tool_call_delta=tool_call_delta,
+                        placement=_current_placement(),
+                        parsers=arg_parsers,
+                    )
 
         # Flush any tail text buffered while checking for split "<function_calls" markers.
         filtered_content_tail = xml_tool_call_content_filter.flush()

backend/onyx/chat/tool_call_args_streaming.py

@@ -0,0 +1,77 @@
+from collections.abc import Generator
+from collections.abc import Mapping
+from typing import Any
+from typing import Type
+
+from onyx.llm.model_response import ChatCompletionDeltaToolCall
+from onyx.server.query_and_chat.placement import Placement
+from onyx.server.query_and_chat.streaming_models import Packet
+from onyx.server.query_and_chat.streaming_models import ToolCallArgumentDelta
+from onyx.tools.built_in_tools import TOOL_NAME_TO_CLASS
+from onyx.tools.interface import Tool
+from onyx.utils.jsonriver import Parser
+
+
+def _get_tool_class(
+    tool_calls_in_progress: Mapping[int, Mapping[str, Any]],
+    tool_call_delta: ChatCompletionDeltaToolCall,
+) -> Type[Tool] | None:
+    """Look up the Tool subclass for a streaming tool call delta."""
+    tool_name = tool_calls_in_progress.get(tool_call_delta.index, {}).get("name")
+    if not tool_name:
+        return None
+    return TOOL_NAME_TO_CLASS.get(tool_name)
+
+
+def maybe_emit_argument_delta(
+    tool_calls_in_progress: Mapping[int, Mapping[str, Any]],
+    tool_call_delta: ChatCompletionDeltaToolCall,
+    placement: Placement,
+    parsers: dict[int, Parser],
+) -> Generator[Packet, None, None]:
+    """Emit decoded tool-call argument deltas to the frontend.
+
+    Uses a ``jsonriver.Parser`` per tool-call index to incrementally parse
+    the JSON argument string and extract only the newly-appended content
+    for each string-valued argument.
+
+    NOTE: Non-string arguments (numbers, booleans, null, arrays, objects)
+    are skipped — they are available in the final tool-call kickoff packet.
+
+    ``parsers`` is a mutable dict keyed by tool-call index. A new
+    ``Parser`` is created automatically for each new index.
+    """
+    tool_cls = _get_tool_class(tool_calls_in_progress, tool_call_delta)
+    if not tool_cls or not tool_cls.should_emit_argument_deltas():
+        return
+
+    fn = tool_call_delta.function
+    delta_fragment = fn.arguments if fn else None
+    if not delta_fragment:
+        return
+
+    idx = tool_call_delta.index
+    if idx not in parsers:
+        parsers[idx] = Parser()
+    parser = parsers[idx]
+
+    deltas = parser.feed(delta_fragment)
+
+    argument_deltas: dict[str, str] = {}
+    for delta in deltas:
+        if isinstance(delta, dict):
+            for key, value in delta.items():
+                if isinstance(value, str):
+                    argument_deltas[key] = argument_deltas.get(key, "") + value
+
+    if not argument_deltas:
+        return
+
+    tc_data = tool_calls_in_progress[tool_call_delta.index]
+    yield Packet(
+        placement=placement,
+        obj=ToolCallArgumentDelta(
+            tool_type=tc_data.get("name", ""),
+            argument_deltas=argument_deltas,
+        ),
+    )

backend/onyx/configs/app_configs.py

@@ -68,6 +68,10 @@ FILE_TOKEN_COUNT_THRESHOLD = int(
     os.environ.get("FILE_TOKEN_COUNT_THRESHOLD", str(_DEFAULT_FILE_TOKEN_LIMIT))
 )
 
+# Maximum upload size for a single user file (chat/projects) in MB.
+USER_FILE_MAX_UPLOAD_SIZE_MB = int(os.environ.get("USER_FILE_MAX_UPLOAD_SIZE_MB") or 50)
+USER_FILE_MAX_UPLOAD_SIZE_BYTES = USER_FILE_MAX_UPLOAD_SIZE_MB * 1024 * 1024
+
 # If set to true, will show extra/uncommon connectors in the "Other" category
 SHOW_EXTRA_CONNECTORS = os.environ.get("SHOW_EXTRA_CONNECTORS", "").lower() == "true"
 

backend/onyx/connectors/google_drive/connector.py

@@ -1722,6 +1722,7 @@ class GoogleDriveConnector(
                         primary_admin_email=self.primary_admin_email,
                         google_domain=self.google_domain,
                     ),
+                    retriever_email=file.user_email,
                 ):
                     slim_batch.append(doc)
 

backend/onyx/connectors/google_drive/doc_conversion.py

@@ -476,6 +476,7 @@ def _get_external_access_for_raw_gdrive_file(
     company_domain: str,
     retriever_drive_service: GoogleDriveService | None,
     admin_drive_service: GoogleDriveService,
+    fallback_user_email: str,
     add_prefix: bool = False,
 ) -> ExternalAccess:
     """
@@ -484,6 +485,8 @@ def _get_external_access_for_raw_gdrive_file(
     add_prefix: When True, prefix group IDs with source type (for indexing path).
                When False (default), leave unprefixed (for permission sync path
                where upsert_document_external_perms handles prefixing).
+    fallback_user_email: When permission info can't be retrieved (e.g. externally-owned
+               files), fall back to granting access to this user.
     """
     external_access_fn = cast(
         Callable[
@@ -492,6 +495,7 @@ def _get_external_access_for_raw_gdrive_file(
                 str,
                 GoogleDriveService | None,
                 GoogleDriveService,
+                str,
                 bool,
             ],
             ExternalAccess,
@@ -507,6 +511,7 @@ def _get_external_access_for_raw_gdrive_file(
         company_domain,
         retriever_drive_service,
         admin_drive_service,
+        fallback_user_email,
         add_prefix,
     )
 
@@ -672,6 +677,7 @@ def _convert_drive_item_to_document(
                     creds, user_email=permission_sync_context.primary_admin_email
                 ),
                 add_prefix=True,  # Indexing path - prefix here
+                fallback_user_email=retriever_email,
             )
             if permission_sync_context
             else None
@@ -753,6 +759,7 @@ def build_slim_document(
     # if not specified, we will not sync permissions
     # will also be a no-op if EE is not enabled
     permission_sync_context: PermissionSyncContext | None,
+    retriever_email: str,
 ) -> SlimDocument | None:
     if file.get("mimeType") in [DRIVE_FOLDER_TYPE, DRIVE_SHORTCUT_TYPE]:
         return None
@@ -774,6 +781,7 @@ def build_slim_document(
                 creds,
                 user_email=permission_sync_context.primary_admin_email,
             ),
+            fallback_user_email=retriever_email,
         )
         if permission_sync_context
         else None

backend/onyx/db/models.py

@@ -36,9 +36,11 @@ from sqlalchemy import Text
 from sqlalchemy import text
 from sqlalchemy import UniqueConstraint
 from sqlalchemy.dialects import postgresql
+from sqlalchemy import event
 from sqlalchemy.engine.interfaces import Dialect
 from sqlalchemy.orm import DeclarativeBase
 from sqlalchemy.orm import Mapped
+from sqlalchemy.orm import Mapper
 from sqlalchemy.orm import mapped_column
 from sqlalchemy.orm import relationship
 from sqlalchemy.types import LargeBinary
@@ -117,10 +119,50 @@ class Base(DeclarativeBase):
     __abstract__ = True
 
 
-class EncryptedString(TypeDecorator):
+class _EncryptedBase(TypeDecorator):
+    """Base for encrypted column types that wrap values in SensitiveValue."""
+
     impl = LargeBinary
-    # This type's behavior is fully deterministic and doesn't depend on any external factors.
     cache_ok = True
+    _is_json: bool = False
+
+    def wrap_raw(self, value: Any) -> SensitiveValue:
+        """Encrypt a raw value and wrap it in SensitiveValue.
+
+        Called by the attribute set event so the Python-side type is always
+        SensitiveValue, regardless of whether the value was loaded from the DB
+        or assigned in application code.
+        """
+        if self._is_json:
+            if not isinstance(value, dict):
+                raise TypeError(
+                    f"EncryptedJson column expected dict, got {type(value).__name__}"
+                )
+            raw_str = json.dumps(value)
+        else:
+            if not isinstance(value, str):
+                raise TypeError(
+                    f"EncryptedString column expected str, got {type(value).__name__}"
+                )
+            raw_str = value
+        return SensitiveValue(
+            encrypted_bytes=encrypt_string_to_bytes(raw_str),
+            decrypt_fn=decrypt_bytes_to_string,
+            is_json=self._is_json,
+        )
+
+    def compare_values(self, x: Any, y: Any) -> bool:
+        if x is None or y is None:
+            return x == y
+        if isinstance(x, SensitiveValue):
+            x = x.get_value(apply_mask=False)
+        if isinstance(y, SensitiveValue):
+            y = y.get_value(apply_mask=False)
+        return x == y
+
+
+class EncryptedString(_EncryptedBase):
+    _is_json: bool = False
 
     def process_bind_param(
         self, value: str | SensitiveValue[str] | None, dialect: Dialect  # noqa: ARG002
@@ -144,20 +186,9 @@ class EncryptedString(TypeDecorator):
             )
         return None
 
-    def compare_values(self, x: Any, y: Any) -> bool:
-        if x is None or y is None:
-            return x == y
-        if isinstance(x, SensitiveValue):
-            x = x.get_value(apply_mask=False)
-        if isinstance(y, SensitiveValue):
-            y = y.get_value(apply_mask=False)
-        return x == y
 
-
-class EncryptedJson(TypeDecorator):
-    impl = LargeBinary
-    # This type's behavior is fully deterministic and doesn't depend on any external factors.
-    cache_ok = True
+class EncryptedJson(_EncryptedBase):
+    _is_json: bool = True
 
     def process_bind_param(
         self,
@@ -165,9 +196,7 @@ class EncryptedJson(TypeDecorator):
         dialect: Dialect,  # noqa: ARG002
     ) -> bytes | None:
         if value is not None:
-            # Handle both raw dicts and SensitiveValue wrappers
             if isinstance(value, SensitiveValue):
-                # Get raw value for storage
                 value = value.get_value(apply_mask=False)
             json_str = json.dumps(value)
             return encrypt_string_to_bytes(json_str)
@@ -184,14 +213,40 @@ class EncryptedJson(TypeDecorator):
             )
         return None
 
-    def compare_values(self, x: Any, y: Any) -> bool:
-        if x is None or y is None:
-            return x == y
-        if isinstance(x, SensitiveValue):
-            x = x.get_value(apply_mask=False)
-        if isinstance(y, SensitiveValue):
-            y = y.get_value(apply_mask=False)
-        return x == y
+
+_REGISTERED_ATTRS: set[str] = set()
+
+
+@event.listens_for(Mapper, "mapper_configured")
+def _register_sensitive_value_set_events(
+    mapper: Mapper,
+    class_: type,
+) -> None:
+    """Auto-wrap raw values in SensitiveValue when assigned to encrypted columns."""
+    for prop in mapper.column_attrs:
+        for col in prop.columns:
+            if isinstance(col.type, _EncryptedBase):
+                col_type = col.type
+                attr = getattr(class_, prop.key)
+
+                # Guard against double-registration (e.g. if mapper is
+                # re-configured in test setups)
+                attr_key = f"{class_.__qualname__}.{prop.key}"
+                if attr_key in _REGISTERED_ATTRS:
+                    continue
+                _REGISTERED_ATTRS.add(attr_key)
+
+                @event.listens_for(attr, "set", retval=True)
+                def _wrap_value(
+                    target: Any,  # noqa: ARG001
+                    value: Any,
+                    oldvalue: Any,  # noqa: ARG001
+                    initiator: Any,  # noqa: ARG001
+                    _col_type: _EncryptedBase = col_type,
+                ) -> Any:
+                    if value is not None and not isinstance(value, SensitiveValue):
+                        return _col_type.wrap_raw(value)
+                    return value
 
 
 class NullFilteredString(TypeDecorator):

backend/onyx/db/rotate_encryption_key.py

@@ -0,0 +1,161 @@
+"""Rotate encryption key for all encrypted columns.
+
+Dynamically discovers all columns using EncryptedString / EncryptedJson,
+decrypts each value with the old key, and re-encrypts with the current
+ENCRYPTION_KEY_SECRET.
+
+The operation is idempotent: rows already encrypted with the current key
+are skipped. Commits are made in batches so a crash mid-rotation can be
+safely resumed by re-running.
+"""
+
+import json
+from typing import Any
+
+from sqlalchemy import LargeBinary
+from sqlalchemy import select
+from sqlalchemy import update
+from sqlalchemy.orm import Session
+
+from onyx.configs.app_configs import ENCRYPTION_KEY_SECRET
+from onyx.db.models import Base
+from onyx.db.models import EncryptedJson
+from onyx.db.models import EncryptedString
+from onyx.utils.encryption import decrypt_bytes_to_string
+from onyx.utils.logger import setup_logger
+from onyx.utils.variable_functionality import global_version
+
+logger = setup_logger()
+
+_BATCH_SIZE = 500
+
+
+def _can_decrypt_with_current_key(data: bytes) -> bool:
+    """Check if data is already encrypted with the current key.
+
+    Passes the key explicitly so the fallback-to-raw-decode path in
+    _decrypt_bytes is NOT triggered — a clean success/failure signal.
+    """
+    try:
+        decrypt_bytes_to_string(data, key=ENCRYPTION_KEY_SECRET)
+        return True
+    except Exception:
+        return False
+
+
+def _discover_encrypted_columns() -> list[tuple[type, str, list[str], bool]]:
+    """Walk all ORM models and find columns using EncryptedString/EncryptedJson.
+
+    Returns list of (ModelClass, column_attr_name, [pk_attr_names], is_json).
+    """
+    results: list[tuple[type, str, list[str], bool]] = []
+
+    for mapper in Base.registry.mappers:
+        model_cls = mapper.class_
+        pk_names = [col.key for col in mapper.primary_key]
+
+        for prop in mapper.column_attrs:
+            for col in prop.columns:
+                if isinstance(col.type, EncryptedJson):
+                    results.append((model_cls, prop.key, pk_names, True))
+                elif isinstance(col.type, EncryptedString):
+                    results.append((model_cls, prop.key, pk_names, False))
+
+    return results
+
+
+def rotate_encryption_key(
+    db_session: Session,
+    old_key: str | None,
+    dry_run: bool = False,
+) -> dict[str, int]:
+    """Decrypt all encrypted columns with old_key and re-encrypt with the current key.
+
+    Args:
+        db_session: Active database session.
+        old_key: The previous encryption key. Pass None or "" if values were
+                 not previously encrypted with a key.
+        dry_run: If True, count rows that need rotation without modifying data.
+
+    Returns:
+        Dict of "table.column" -> number of rows re-encrypted (or would be).
+
+    Commits every _BATCH_SIZE rows so that locks are held briefly and progress
+    is preserved on crash. Already-rotated rows are detected and skipped,
+    making the operation safe to re-run.
+    """
+    if not global_version.is_ee_version():
+        raise RuntimeError("EE mode is not enabled — rotation requires EE encryption.")
+
+    if not ENCRYPTION_KEY_SECRET:
+        raise RuntimeError(
+            "ENCRYPTION_KEY_SECRET is not set — cannot rotate. "
+            "Set the target encryption key in the environment before running."
+        )
+
+    encrypted_columns = _discover_encrypted_columns()
+    totals: dict[str, int] = {}
+
+    for model_cls, col_name, pk_names, is_json in encrypted_columns:
+        table_name: str = model_cls.__tablename__  # type: ignore[attr-defined]
+        col_attr = getattr(model_cls, col_name)
+        pk_attrs = [getattr(model_cls, pk) for pk in pk_names]
+
+        # Read raw bytes directly, bypassing the TypeDecorator
+        raw_col = col_attr.property.columns[0]
+
+        stmt = select(*pk_attrs, raw_col.cast(LargeBinary)).where(col_attr.is_not(None))
+        rows = db_session.execute(stmt).all()
+
+        reencrypted = 0
+        batch_pending = 0
+        for row in rows:
+            raw_bytes: bytes | None = row[-1]
+            if raw_bytes is None:
+                continue
+
+            if _can_decrypt_with_current_key(raw_bytes):
+                continue
+
+            try:
+                if not old_key:
+                    decrypted_str = raw_bytes.decode("utf-8")
+                else:
+                    decrypted_str = decrypt_bytes_to_string(raw_bytes, key=old_key)
+
+                # For EncryptedJson, parse back to dict so the TypeDecorator
+                # can json.dumps() it cleanly (avoids double-encoding).
+                value: Any = json.loads(decrypted_str) if is_json else decrypted_str
+            except (ValueError, UnicodeDecodeError) as e:
+                pk_vals = [row[i] for i in range(len(pk_names))]
+                logger.warning(
+                    f"Could not decrypt/parse {table_name}.{col_name} "
+                    f"row {pk_vals} — skipping: {e}"
+                )
+                continue
+
+            if not dry_run:
+                pk_filters = [pk_attr == row[i] for i, pk_attr in enumerate(pk_attrs)]
+                update_stmt = (
+                    update(model_cls).where(*pk_filters).values({col_name: value})
+                )
+                db_session.execute(update_stmt)
+                batch_pending += 1
+
+                if batch_pending >= _BATCH_SIZE:
+                    db_session.commit()
+                    batch_pending = 0
+            reencrypted += 1
+
+        # Flush remaining rows in this column
+        if batch_pending > 0:
+            db_session.commit()
+
+        if reencrypted > 0:
+            totals[f"{table_name}.{col_name}"] = reencrypted
+            logger.info(
+                f"{'[DRY RUN] Would re-encrypt' if dry_run else 'Re-encrypted'} "
+                f"{reencrypted} value(s) in {table_name}.{col_name}"
+            )
+
+    return totals

backend/onyx/mcp_server/tools/search.py

@@ -10,6 +10,7 @@ from onyx.mcp_server.utils import get_indexed_sources
 from onyx.mcp_server.utils import require_access_token
 from onyx.utils.logger import setup_logger
 from onyx.utils.variable_functionality import build_api_server_url_for_http_requests
+from onyx.utils.variable_functionality import global_version
 
 logger = setup_logger()
 
@@ -26,6 +27,14 @@ async def search_indexed_documents(
     Use this tool for information that is not public knowledge and specific to the user,
     their team, their work, or their organization/company.
 
+    Note: In CE mode, this tool uses the chat endpoint internally which invokes an LLM
+    on every call, consuming tokens and adding latency.
+    Additionally, CE callers receive a truncated snippet (blurb) instead of a full document chunk,
+    but this should still be sufficient for most use cases. CE mode functionality should be swapped
+    when a dedicated CE search endpoint is implemented.
+
+    In EE mode, the dedicated search endpoint is used instead.
+
     To find a list of available sources, use the `indexed_sources` resource.
     Returns chunks of text as search results with snippets, scores, and metadata.
 
@@ -111,48 +120,73 @@ async def search_indexed_documents(
         if time_cutoff_dt:
             filters["time_cutoff"] = time_cutoff_dt.isoformat()
 
-    # Build the search request using the new SendSearchQueryRequest format
-    search_request = {
-        "search_query": query,
-        "filters": filters,
-        "num_docs_fed_to_llm_selection": limit,
-        "run_query_expansion": False,
-        "include_content": True,
-        "stream": False,
-    }
+    is_ee = global_version.is_ee_version()
+    base_url = build_api_server_url_for_http_requests(respect_env_override_if_set=True)
+    auth_headers = {"Authorization": f"Bearer {access_token.token}"}
+
+    search_request: dict[str, Any]
+    if is_ee:
+        # EE: use the dedicated search endpoint (no LLM invocation)
+        search_request = {
+            "search_query": query,
+            "filters": filters,
+            "num_docs_fed_to_llm_selection": limit,
+            "run_query_expansion": False,
+            "include_content": True,
+            "stream": False,
+        }
+        endpoint = f"{base_url}/search/send-search-message"
+        error_key = "error"
+        docs_key = "search_docs"
+        content_field = "content"
+    else:
+        # CE: fall back to the chat endpoint (invokes LLM, consumes tokens)
+        search_request = {
+            "message": query,
+            "stream": False,
+            "chat_session_info": {},
+        }
+        if filters:
+            search_request["internal_search_filters"] = filters
+        endpoint = f"{base_url}/chat/send-chat-message"
+        error_key = "error_msg"
+        docs_key = "top_documents"
+        content_field = "blurb"
 
-    # Call the API server using the new send-search-message route
     try:
         response = await get_http_client().post(
-            f"{build_api_server_url_for_http_requests(respect_env_override_if_set=True)}/search/send-search-message",
+            endpoint,
             json=search_request,
-            headers={"Authorization": f"Bearer {access_token.token}"},
+            headers=auth_headers,
         )
         response.raise_for_status()
         result = response.json()
 
         # Check for error in response
-        if result.get("error"):
+        if result.get(error_key):
             return {
                 "documents": [],
                 "total_results": 0,
                 "query": query,
-                "error": result.get("error"),
+                "error": result.get(error_key),
             }
 
-        # Return simplified format for MCP clients
-        fields_to_return = [
-            "semantic_identifier",
-            "content",
-            "source_type",
-            "link",
-            "score",
-        ]
         documents = [
-            {key: doc.get(key) for key in fields_to_return}
-            for doc in result.get("search_docs", [])
+            {
+                "semantic_identifier": doc.get("semantic_identifier"),
+                "content": doc.get(content_field),
+                "source_type": doc.get("source_type"),
+                "link": doc.get("link"),
+                "score": doc.get("score"),
+            }
+            for doc in result.get(docs_key, [])
         ]
 
+        # NOTE: search depth is controlled by the backend persona defaults, not `limit`.
+        # `limit` only caps the returned list; fewer results may be returned if the
+        # backend retrieves fewer documents than requested.
+        documents = documents[:limit]
+
         logger.info(
             f"Onyx MCP Server: Internal search returned {len(documents)} results"
         )
@@ -160,7 +194,6 @@ async def search_indexed_documents(
             "documents": documents,
             "total_results": len(documents),
             "query": query,
-            "executed_queries": result.get("all_executed_queries", [query]),
         }
     except Exception as e:
         logger.error(f"Onyx MCP Server: Document search error: {e}", exc_info=True)

backend/onyx/server/features/projects/projects_file_utils.py

@@ -10,6 +10,8 @@ from pydantic import Field
 from sqlalchemy.orm import Session
 
 from onyx.configs.app_configs import FILE_TOKEN_COUNT_THRESHOLD
+from onyx.configs.app_configs import USER_FILE_MAX_UPLOAD_SIZE_BYTES
+from onyx.configs.app_configs import USER_FILE_MAX_UPLOAD_SIZE_MB
 from onyx.db.llm import fetch_default_llm_model
 from onyx.file_processing.extract_file_text import extract_file_text
 from onyx.file_processing.extract_file_text import get_file_ext
@@ -35,6 +37,38 @@ def get_safe_filename(upload: UploadFile) -> str:
     return upload.filename
 
 
+def get_upload_size_bytes(upload: UploadFile) -> int | None:
+    """Best-effort file size in bytes without consuming the stream."""
+    if upload.size is not None:
+        return upload.size
+
+    try:
+        current_pos = upload.file.tell()
+        upload.file.seek(0, 2)
+        size = upload.file.tell()
+        upload.file.seek(current_pos)
+        return size
+    except Exception as e:
+        logger.warning(
+            "Could not determine upload size via stream seek "
+            f"(filename='{get_safe_filename(upload)}', "
+            f"error_type={type(e).__name__}, error={e})"
+        )
+        return None
+
+
+def is_upload_too_large(upload: UploadFile, max_bytes: int) -> bool:
+    """Return True when upload size is known and exceeds max_bytes."""
+    size_bytes = get_upload_size_bytes(upload)
+    if size_bytes is None:
+        logger.warning(
+            "Could not determine upload size; skipping size-limit check for "
+            f"'{get_safe_filename(upload)}'"
+        )
+        return False
+    return size_bytes > max_bytes
+
+
 # Guard against extremely large images
 Image.MAX_IMAGE_PIXELS = 12000 * 12000
 
@@ -159,6 +193,18 @@ def categorize_uploaded_files(
     for upload in files:
         try:
             filename = get_safe_filename(upload)
+
+            # Size limit is a hard safety cap and is enforced even when token
+            # threshold checks are skipped via SKIP_USERFILE_THRESHOLD settings.
+            if is_upload_too_large(upload, USER_FILE_MAX_UPLOAD_SIZE_BYTES):
+                results.rejected.append(
+                    RejectedFile(
+                        filename=filename,
+                        reason=f"Exceeds {USER_FILE_MAX_UPLOAD_SIZE_MB} MB file size limit",
+                    )
+                )
+                continue
+
             extension = get_file_ext(filename)
 
             # If image, estimate tokens via dedicated method first

backend/onyx/server/query_and_chat/streaming_models.py

@@ -41,6 +41,7 @@ class StreamingType(Enum):
     REASONING_DONE = "reasoning_done"
     CITATION_INFO = "citation_info"
     TOOL_CALL_DEBUG = "tool_call_debug"
+    TOOL_CALL_ARGUMENT_DELTA = "tool_call_argument_delta"
 
     MEMORY_TOOL_START = "memory_tool_start"
     MEMORY_TOOL_DELTA = "memory_tool_delta"
@@ -259,6 +260,15 @@ class CustomToolDelta(BaseObj):
     file_ids: list[str] | None = None
 
 
+class ToolCallArgumentDelta(BaseObj):
+    type: Literal["tool_call_argument_delta"] = (
+        StreamingType.TOOL_CALL_ARGUMENT_DELTA.value
+    )
+
+    tool_type: str
+    argument_deltas: dict[str, Any]
+
+
 ################################################
 # File Reader Packets
 ################################################
@@ -379,6 +389,7 @@ PacketObj = Union[
     # Citation Packets
     CitationInfo,
     ToolCallDebug,
+    ToolCallArgumentDelta,
     # Deep Research Packets
     DeepResearchPlanStart,
     DeepResearchPlanDelta,

backend/onyx/server/settings/models.py

@@ -78,6 +78,7 @@ class Settings(BaseModel):
 
     # User Knowledge settings
     user_knowledge_enabled: bool | None = True
+    user_file_max_upload_size_mb: int | None = None
 
     # Connector settings
     show_extra_connectors: bool | None = True

backend/onyx/server/settings/store.py

@@ -3,6 +3,7 @@ from onyx.configs.app_configs import DISABLE_USER_KNOWLEDGE
 from onyx.configs.app_configs import ENABLE_OPENSEARCH_INDEXING_FOR_ONYX
 from onyx.configs.app_configs import ONYX_QUERY_HISTORY_TYPE
 from onyx.configs.app_configs import SHOW_EXTRA_CONNECTORS
+from onyx.configs.app_configs import USER_FILE_MAX_UPLOAD_SIZE_MB
 from onyx.configs.constants import KV_SETTINGS_KEY
 from onyx.configs.constants import OnyxRedisLocks
 from onyx.key_value_store.factory import get_kv_store
@@ -50,6 +51,7 @@ def load_settings() -> Settings:
     if DISABLE_USER_KNOWLEDGE:
         settings.user_knowledge_enabled = False
 
+    settings.user_file_max_upload_size_mb = USER_FILE_MAX_UPLOAD_SIZE_MB
     settings.show_extra_connectors = SHOW_EXTRA_CONNECTORS
     settings.opensearch_indexing_enabled = ENABLE_OPENSEARCH_INDEXING_FOR_ONYX
     return settings

backend/onyx/tools/built_in_tools.py

@@ -56,3 +56,23 @@ def get_built_in_tool_ids() -> list[str]:
 
 def get_built_in_tool_by_id(in_code_tool_id: str) -> Type[BUILT_IN_TOOL_TYPES]:
     return BUILT_IN_TOOL_MAP[in_code_tool_id]
+
+
+def _build_tool_name_to_class() -> dict[str, Type[BUILT_IN_TOOL_TYPES]]:
+    """Build a mapping from LLM-facing tool name to tool class."""
+    result: dict[str, Type[BUILT_IN_TOOL_TYPES]] = {}
+    for cls in BUILT_IN_TOOL_MAP.values():
+        name_attr = cls.__dict__.get("name")
+        if isinstance(name_attr, property) and name_attr.fget is not None:
+            tool_name = name_attr.fget(cls)
+        elif isinstance(name_attr, str):
+            tool_name = name_attr
+        else:
+            raise ValueError(
+                f"Built-in tool {cls.__name__} must define a valid LLM-facing tool name"
+            )
+        result[tool_name] = cls
+    return result
+
+
+TOOL_NAME_TO_CLASS: dict[str, Type[BUILT_IN_TOOL_TYPES]] = _build_tool_name_to_class()

backend/onyx/tools/interface.py

@@ -92,3 +92,7 @@ class Tool(abc.ABC, Generic[TOverride]):
         **llm_kwargs: Any,
     ) -> ToolResponse:
         raise NotImplementedError
+
+    @classmethod
+    def should_emit_argument_deltas(cls) -> bool:
+        return False

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

@@ -376,3 +376,8 @@ class PythonTool(Tool[PythonToolOverrideKwargs]):
                     rich_response=None,
                     llm_facing_response=llm_response,
                 )
+
+    @classmethod
+    @override
+    def should_emit_argument_deltas(cls) -> bool:
+        return True

backend/onyx/utils/encryption.py

@@ -11,16 +11,20 @@ logger = setup_logger()
 
 
 # IMPORTANT DO NOT DELETE, THIS IS USED BY fetch_versioned_implementation
-def _encrypt_string(input_str: str) -> bytes:
+def _encrypt_string(input_str: str, key: str | None = None) -> bytes:  # noqa: ARG001
     if ENCRYPTION_KEY_SECRET:
         logger.warning("MIT version of Onyx does not support encryption of secrets.")
+    elif key is not None:
+        logger.debug("MIT encrypt called with explicit key — key ignored.")
     return input_str.encode()
 
 
 # IMPORTANT DO NOT DELETE, THIS IS USED BY fetch_versioned_implementation
-def _decrypt_bytes(input_bytes: bytes) -> str:
-    # No need to double warn. If you wish to learn more about encryption features
-    # refer to the Onyx EE code
+def _decrypt_bytes(input_bytes: bytes, key: str | None = None) -> str:  # noqa: ARG001
+    if ENCRYPTION_KEY_SECRET:
+        logger.warning("MIT version of Onyx does not support decryption of secrets.")
+    elif key is not None:
+        logger.debug("MIT decrypt called with explicit key — key ignored.")
     return input_bytes.decode()
 
 
@@ -86,15 +90,15 @@ def _mask_list(items: list[Any]) -> list[Any]:
     return masked
 
 
-def encrypt_string_to_bytes(intput_str: str) -> bytes:
+def encrypt_string_to_bytes(intput_str: str, key: str | None = None) -> bytes:
     versioned_encryption_fn = fetch_versioned_implementation(
         "onyx.utils.encryption", "_encrypt_string"
     )
-    return versioned_encryption_fn(intput_str)
+    return versioned_encryption_fn(intput_str, key=key)
 
 
-def decrypt_bytes_to_string(intput_bytes: bytes) -> str:
+def decrypt_bytes_to_string(intput_bytes: bytes, key: str | None = None) -> str:
     versioned_decryption_fn = fetch_versioned_implementation(
         "onyx.utils.encryption", "_decrypt_bytes"
     )
-    return versioned_decryption_fn(intput_bytes)
+    return versioned_decryption_fn(intput_bytes, key=key)

backend/onyx/utils/sensitive.py

@@ -128,6 +128,8 @@ class SensitiveValue(Generic[T]):
         value = self._decrypt()
 
         if not apply_mask:
+            # Callers must not mutate the returned dict — doing so would
+            # desync the cache from the encrypted bytes and the DB.
             return value
 
         # Apply masking
@@ -174,18 +176,20 @@ class SensitiveValue(Generic[T]):
         )
 
     def __eq__(self, other: Any) -> bool:
-        """Prevent direct comparison which might expose value."""
-        if isinstance(other, SensitiveValue):
-            # Compare encrypted bytes for equality check
-            return self._encrypted_bytes == other._encrypted_bytes
-        raise SensitiveAccessError(
-            "Cannot compare SensitiveValue with non-SensitiveValue. "
-            "Use .get_value(apply_mask=True/False) to access the value for comparison."
-        )
+        """Compare SensitiveValues by their decrypted content."""
+        # NOTE: if you attempt to compare a string/dict to a SensitiveValue,
+        # this comparison will return NotImplemented, which then evaluates to False.
+        # This is the convention and required for SQLAlchemy's attribute tracking.
+        if not isinstance(other, SensitiveValue):
+            return NotImplemented
+        return self._decrypt() == other._decrypt()
 
     def __hash__(self) -> int:
-        """Allow hashing based on encrypted bytes."""
-        return hash(self._encrypted_bytes)
+        """Hash based on decrypted content."""
+        value = self._decrypt()
+        if isinstance(value, dict):
+            return hash(json.dumps(value, sort_keys=True))
+        return hash(value)
 
     # Prevent JSON serialization
     def __json__(self) -> Any:

backend/scripts/decrypt.py

@@ -1,48 +1,93 @@
+"""Decrypt a raw hex-encoded credential value.
+
+Usage:
+    python -m scripts.decrypt <hex_value>
+    python -m scripts.decrypt <hex_value> --key "my-encryption-key"
+    python -m scripts.decrypt <hex_value> --key ""
+
+Pass --key "" to skip decryption and just decode the raw bytes as UTF-8.
+Omit --key to use the current ENCRYPTION_KEY_SECRET from the environment.
+"""
+
+import argparse
 import binascii
 import json
+import os
 import sys
 
-from onyx.utils.encryption import decrypt_bytes_to_string
+parent_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.append(parent_dir)
+
+from onyx.utils.encryption import decrypt_bytes_to_string  # noqa: E402
+from onyx.utils.variable_functionality import global_version  # noqa: E402
 
 
-def decrypt_raw_credential(encrypted_value: str) -> None:
-    """Decrypt and display a raw encrypted credential value
+def decrypt_raw_credential(encrypted_value: str, key: str | None = None) -> None:
+    """Decrypt and display a raw encrypted credential value.
 
     Args:
-        encrypted_value: The hex encoded encrypted credential value
+        encrypted_value: The hex-encoded encrypted credential value.
+        key: Encryption key to use. None means use ENCRYPTION_KEY_SECRET,
+             empty string means just decode as UTF-8.
     """
+    # Strip common hex prefixes
+    if encrypted_value.startswith("\\x"):
+        encrypted_value = encrypted_value[2:]
+    elif encrypted_value.startswith("x"):
+        encrypted_value = encrypted_value[1:]
+    print(encrypted_value)
+
     try:
-        # If string starts with 'x', remove it as it's just a prefix indicating hex
-        if encrypted_value.startswith("x"):
-            encrypted_value = encrypted_value[1:]
-        elif encrypted_value.startswith("\\x"):
-            encrypted_value = encrypted_value[2:]
-
-        # Convert hex string to bytes
-        encrypted_bytes = binascii.unhexlify(encrypted_value)
-
-        # Decrypt the bytes
-        decrypted_str = decrypt_bytes_to_string(encrypted_bytes)
-
-        # Parse and pretty print the decrypted JSON
-        decrypted_json = json.loads(decrypted_str)
-        print("Decrypted credential value:")
-        print(json.dumps(decrypted_json, indent=2))
-
+        raw_bytes = binascii.unhexlify(encrypted_value)
     except binascii.Error:
-        print("Error: Invalid hex encoded string")
+        print("Error: Invalid hex-encoded string")
+        sys.exit(1)
 
-    except json.JSONDecodeError as e:
-        print(f"Decrypted raw value (not JSON): {e}")
+    if key == "":
+        # Empty key → just decode as UTF-8, no decryption
+        try:
+            decrypted_str = raw_bytes.decode("utf-8")
+        except UnicodeDecodeError as e:
+            print(f"Error decoding bytes as UTF-8: {e}")
+            sys.exit(1)
+    else:
+        print(key)
+        try:
+            decrypted_str = decrypt_bytes_to_string(raw_bytes, key=key)
+        except Exception as e:
+            print(f"Error decrypting value: {e}")
+            sys.exit(1)
 
-    except Exception as e:
-        print(f"Error decrypting value: {e}")
+    # Try to pretty-print as JSON, otherwise print raw
+    try:
+        parsed = json.loads(decrypted_str)
+        print(json.dumps(parsed, indent=2))
+    except json.JSONDecodeError:
+        print(decrypted_str)
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(
+        description="Decrypt a hex-encoded credential value."
+    )
+    parser.add_argument(
+        "value",
+        help="Hex-encoded encrypted value to decrypt.",
+    )
+    parser.add_argument(
+        "--key",
+        default=None,
+        help=(
+            "Encryption key. Omit to use ENCRYPTION_KEY_SECRET from env. "
+            'Pass "" (empty) to just decode as UTF-8 without decryption.'
+        ),
+    )
+    args = parser.parse_args()
+
+    global_version.set_ee()
+    decrypt_raw_credential(args.value, key=args.key)
+    global_version.unset_ee()
 
 
 if __name__ == "__main__":
-    if len(sys.argv) != 2:
-        print("Usage: python decrypt.py <hex_encoded_encrypted_value>")
-        sys.exit(1)
-
-    encrypted_value = sys.argv[1]
-    decrypt_raw_credential(encrypted_value)
+    main()

backend/scripts/reencrypt_secrets.py

@@ -0,0 +1,107 @@
+"""Re-encrypt secrets under the current ENCRYPTION_KEY_SECRET.
+
+Decrypts all encrypted columns using the old key (or raw decode if the old key
+is empty), then re-encrypts them with the current ENCRYPTION_KEY_SECRET.
+
+Usage (docker):
+    docker exec -it onyx-api_server-1 \
+        python -m scripts.reencrypt_secrets --old-key "previous-key"
+
+Usage (kubernetes):
+    kubectl exec -it <pod> -- \
+        python -m scripts.reencrypt_secrets --old-key "previous-key"
+
+Omit --old-key (or pass "") if secrets were not previously encrypted.
+
+For multi-tenant deployments, pass --tenant-id to target a specific tenant,
+or --all-tenants to iterate every tenant.
+"""
+
+import argparse
+import os
+import sys
+
+parent_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+sys.path.append(parent_dir)
+
+from onyx.db.rotate_encryption_key import rotate_encryption_key  # noqa: E402
+from onyx.db.engine.sql_engine import get_session_with_tenant  # noqa: E402
+from onyx.db.engine.sql_engine import SqlEngine  # noqa: E402
+from onyx.db.engine.tenant_utils import get_all_tenant_ids  # noqa: E402
+from onyx.utils.variable_functionality import global_version  # noqa: E402
+from shared_configs.configs import POSTGRES_DEFAULT_SCHEMA  # noqa: E402
+
+
+def _run_for_tenant(tenant_id: str, old_key: str | None, dry_run: bool = False) -> None:
+    print(f"Re-encrypting secrets for tenant: {tenant_id}")
+    with get_session_with_tenant(tenant_id=tenant_id) as db_session:
+        results = rotate_encryption_key(db_session, old_key=old_key, dry_run=dry_run)
+
+    if results:
+        for col, count in results.items():
+            print(
+                f"  {col}: {count} row(s) {'would be ' if dry_run else ''}re-encrypted"
+            )
+    else:
+        print("No rows needed re-encryption.")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(
+        description="Re-encrypt secrets under the current encryption key."
+    )
+    parser.add_argument(
+        "--old-key",
+        default=None,
+        help="Previous encryption key. Omit or pass empty string if not applicable.",
+    )
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+        help="Show what would be re-encrypted without making changes.",
+    )
+
+    tenant_group = parser.add_mutually_exclusive_group()
+    tenant_group.add_argument(
+        "--tenant-id",
+        default=None,
+        help="Target a specific tenant schema.",
+    )
+    tenant_group.add_argument(
+        "--all-tenants",
+        action="store_true",
+        help="Iterate all tenants.",
+    )
+
+    args = parser.parse_args()
+
+    old_key = args.old_key if args.old_key else None
+
+    global_version.set_ee()
+    SqlEngine.init_engine(pool_size=5, max_overflow=2)
+
+    if args.dry_run:
+        print("DRY RUN — no changes will be made")
+
+    if args.all_tenants:
+        tenant_ids = get_all_tenant_ids()
+        print(f"Found {len(tenant_ids)} tenant(s)")
+        failed_tenants: list[str] = []
+        for tid in tenant_ids:
+            try:
+                _run_for_tenant(tid, old_key, dry_run=args.dry_run)
+            except Exception as e:
+                print(f"  ERROR for tenant {tid}: {e}")
+                failed_tenants.append(tid)
+        if failed_tenants:
+            print(f"FAILED tenants ({len(failed_tenants)}): {failed_tenants}")
+            sys.exit(1)
+    else:
+        tenant_id = args.tenant_id or POSTGRES_DEFAULT_SCHEMA
+        _run_for_tenant(tenant_id, old_key, dry_run=args.dry_run)
+
+    print("Done.")
+
+
+if __name__ == "__main__":
+    main()

backend/tests/external_dependency_unit/db/test_credential_sensitive_value.py

@@ -0,0 +1,90 @@
+"""Test that Credential with nested JSON round-trips through SensitiveValue correctly.
+
+Exercises the full encrypt → store → read → decrypt → SensitiveValue path
+with realistic nested OAuth credential data, and verifies SQLAlchemy dirty
+tracking works with nested dict comparison.
+
+Requires a running Postgres instance.
+"""
+
+from sqlalchemy.orm import Session
+
+from onyx.configs.constants import DocumentSource
+from onyx.db.models import Credential
+from onyx.utils.sensitive import SensitiveValue
+
+# NOTE: this is not the real shape of a Drive credential,
+# but it is intended to test nested JSON credential handling
+
+_NESTED_CRED_JSON = {
+    "oauth_tokens": {
+        "access_token": "ya29.abc123",
+        "refresh_token": "1//xEg-def456",
+    },
+    "scopes": ["read", "write", "admin"],
+    "client_config": {
+        "client_id": "123.apps.googleusercontent.com",
+        "client_secret": "GOCSPX-secret",
+    },
+}
+
+
+def test_nested_credential_json_round_trip(db_session: Session) -> None:
+    """Nested OAuth credential survives encrypt → store → read → decrypt."""
+    credential = Credential(
+        source=DocumentSource.GOOGLE_DRIVE,
+        credential_json=_NESTED_CRED_JSON,
+    )
+    db_session.add(credential)
+    db_session.flush()
+
+    # Immediate read (no DB round-trip) — tests the set event wrapping
+    assert isinstance(credential.credential_json, SensitiveValue)
+    assert credential.credential_json.get_value(apply_mask=False) == _NESTED_CRED_JSON
+
+    # DB round-trip — tests process_result_value
+    db_session.expire(credential)
+    reloaded = credential.credential_json
+    assert isinstance(reloaded, SensitiveValue)
+    assert reloaded.get_value(apply_mask=False) == _NESTED_CRED_JSON
+
+    db_session.rollback()
+
+
+def test_reassign_same_nested_json_not_dirty(db_session: Session) -> None:
+    """Re-assigning the same nested dict should not mark the session dirty."""
+    credential = Credential(
+        source=DocumentSource.GOOGLE_DRIVE,
+        credential_json=_NESTED_CRED_JSON,
+    )
+    db_session.add(credential)
+    db_session.flush()
+
+    # Clear dirty state from the insert
+    db_session.expire(credential)
+    _ = credential.credential_json  # force reload
+
+    # Re-assign identical value
+    credential.credential_json = _NESTED_CRED_JSON  # type: ignore[assignment]
+    assert not db_session.is_modified(credential)
+
+    db_session.rollback()
+
+
+def test_assign_different_nested_json_is_dirty(db_session: Session) -> None:
+    """Assigning a different nested dict should mark the session dirty."""
+    credential = Credential(
+        source=DocumentSource.GOOGLE_DRIVE,
+        credential_json=_NESTED_CRED_JSON,
+    )
+    db_session.add(credential)
+    db_session.flush()
+
+    db_session.expire(credential)
+    _ = credential.credential_json  # force reload
+
+    modified_cred = {**_NESTED_CRED_JSON, "scopes": ["read"]}
+    credential.credential_json = modified_cred  # type: ignore[assignment]
+    assert db_session.is_modified(credential)
+
+    db_session.rollback()

backend/tests/external_dependency_unit/db/test_rotate_encryption_key.py

@@ -0,0 +1,305 @@
+"""Tests for rotate_encryption_key against real Postgres.
+
+Uses real ORM models (Credential, InternetSearchProvider) and the actual
+Postgres database. Discovery is mocked in rotation tests to scope mutations
+to only the test rows — the real _discover_encrypted_columns walk is tested
+separately in TestDiscoverEncryptedColumns.
+
+Requires a running Postgres instance. Run with::
+
+    python -m dotenv -f .vscode/.env run -- pytest tests/external_dependency_unit/db/test_rotate_encryption_key.py
+"""
+
+import json
+from collections.abc import Generator
+from unittest.mock import patch
+
+import pytest
+from sqlalchemy import LargeBinary
+from sqlalchemy import select
+from sqlalchemy import text
+from sqlalchemy.orm import Session
+
+from ee.onyx.utils.encryption import _decrypt_bytes
+from ee.onyx.utils.encryption import _encrypt_string
+from ee.onyx.utils.encryption import _get_trimmed_key
+from onyx.configs.constants import DocumentSource
+from onyx.db.models import Credential
+from onyx.db.models import EncryptedJson
+from onyx.db.models import EncryptedString
+from onyx.db.models import InternetSearchProvider
+from onyx.db.rotate_encryption_key import _discover_encrypted_columns
+from onyx.db.rotate_encryption_key import rotate_encryption_key
+from onyx.utils.variable_functionality import fetch_versioned_implementation
+from onyx.utils.variable_functionality import global_version
+
+EE_MODULE = "ee.onyx.utils.encryption"
+ROTATE_MODULE = "onyx.db.rotate_encryption_key"
+
+OLD_KEY = "o" * 16
+NEW_KEY = "n" * 16
+
+
+@pytest.fixture(autouse=True)
+def _enable_ee() -> Generator[None, None, None]:
+    prev = global_version._is_ee
+    global_version.set_ee()
+    fetch_versioned_implementation.cache_clear()
+    yield
+    global_version._is_ee = prev
+    fetch_versioned_implementation.cache_clear()
+
+
+@pytest.fixture(autouse=True)
+def _clear_key_cache() -> None:
+    _get_trimmed_key.cache_clear()
+
+
+def _raw_credential_bytes(db_session: Session, credential_id: int) -> bytes | None:
+    """Read raw bytes from credential_json, bypassing the TypeDecorator."""
+    col = Credential.__table__.c.credential_json
+    stmt = select(col.cast(LargeBinary)).where(
+        Credential.__table__.c.id == credential_id
+    )
+    return db_session.execute(stmt).scalar()
+
+
+def _raw_isp_bytes(db_session: Session, isp_id: int) -> bytes | None:
+    """Read raw bytes from InternetSearchProvider.api_key."""
+    col = InternetSearchProvider.__table__.c.api_key
+    stmt = select(col.cast(LargeBinary)).where(
+        InternetSearchProvider.__table__.c.id == isp_id
+    )
+    return db_session.execute(stmt).scalar()
+
+
+class TestDiscoverEncryptedColumns:
+    """Verify _discover_encrypted_columns finds real production models."""
+
+    def test_discovers_credential_json(self) -> None:
+        results = _discover_encrypted_columns()
+        found = {
+            (model_cls.__tablename__, col_name, is_json)  # type: ignore[attr-defined]
+            for model_cls, col_name, _, is_json in results
+        }
+        assert ("credential", "credential_json", True) in found
+
+    def test_discovers_internet_search_provider_api_key(self) -> None:
+        results = _discover_encrypted_columns()
+        found = {
+            (model_cls.__tablename__, col_name, is_json)  # type: ignore[attr-defined]
+            for model_cls, col_name, _, is_json in results
+        }
+        assert ("internet_search_provider", "api_key", False) in found
+
+    def test_all_encrypted_string_columns_are_not_json(self) -> None:
+        results = _discover_encrypted_columns()
+        for model_cls, col_name, _, is_json in results:
+            col = getattr(model_cls, col_name).property.columns[0]
+            if isinstance(col.type, EncryptedString):
+                assert not is_json, (
+                    f"{model_cls.__tablename__}.{col_name} is EncryptedString "  # type: ignore[attr-defined]
+                    f"but is_json={is_json}"
+                )
+
+    def test_all_encrypted_json_columns_are_json(self) -> None:
+        results = _discover_encrypted_columns()
+        for model_cls, col_name, _, is_json in results:
+            col = getattr(model_cls, col_name).property.columns[0]
+            if isinstance(col.type, EncryptedJson):
+                assert is_json, (
+                    f"{model_cls.__tablename__}.{col_name} is EncryptedJson "  # type: ignore[attr-defined]
+                    f"but is_json={is_json}"
+                )
+
+
+class TestRotateCredential:
+    """Test rotation against the real Credential table (EncryptedJson).
+
+    Discovery is scoped to only the Credential model to avoid mutating
+    other tables in the test database.
+    """
+
+    @pytest.fixture(autouse=True)
+    def _limit_discovery(self) -> Generator[None, None, None]:
+        with patch(
+            f"{ROTATE_MODULE}._discover_encrypted_columns",
+            return_value=[(Credential, "credential_json", ["id"], True)],
+        ):
+            yield
+
+    @pytest.fixture()
+    def credential_id(
+        self, db_session: Session, tenant_context: None  # noqa: ARG002
+    ) -> Generator[int, None, None]:
+        """Insert a Credential row with raw encrypted bytes, clean up after."""
+        config = {"api_key": "sk-test-1234", "endpoint": "https://example.com"}
+        encrypted = _encrypt_string(json.dumps(config), key=OLD_KEY)
+
+        result = db_session.execute(
+            text(
+                "INSERT INTO credential "
+                "(source, credential_json, admin_public, curator_public) "
+                "VALUES (:source, :cred_json, true, false) "
+                "RETURNING id"
+            ),
+            {"source": DocumentSource.INGESTION_API.value, "cred_json": encrypted},
+        )
+        cred_id = result.scalar_one()
+        db_session.commit()
+
+        yield cred_id
+
+        db_session.execute(
+            text("DELETE FROM credential WHERE id = :id"), {"id": cred_id}
+        )
+        db_session.commit()
+
+    def test_rotates_credential_json(
+        self, db_session: Session, credential_id: int
+    ) -> None:
+        with (
+            patch(f"{ROTATE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+            patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+        ):
+            totals = rotate_encryption_key(db_session, old_key=OLD_KEY)
+
+        assert totals.get("credential.credential_json", 0) >= 1
+
+        raw = _raw_credential_bytes(db_session, credential_id)
+        assert raw is not None
+        decrypted = json.loads(_decrypt_bytes(raw, key=NEW_KEY))
+        assert decrypted["api_key"] == "sk-test-1234"
+        assert decrypted["endpoint"] == "https://example.com"
+
+    def test_skips_already_rotated(
+        self, db_session: Session, credential_id: int
+    ) -> None:
+        with (
+            patch(f"{ROTATE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+            patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+        ):
+            rotate_encryption_key(db_session, old_key=OLD_KEY)
+            _ = rotate_encryption_key(db_session, old_key=OLD_KEY)
+
+        raw = _raw_credential_bytes(db_session, credential_id)
+        assert raw is not None
+        decrypted = json.loads(_decrypt_bytes(raw, key=NEW_KEY))
+        assert decrypted["api_key"] == "sk-test-1234"
+
+    def test_dry_run_does_not_modify(
+        self, db_session: Session, credential_id: int
+    ) -> None:
+        original = _raw_credential_bytes(db_session, credential_id)
+
+        with (
+            patch(f"{ROTATE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+            patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+        ):
+            totals = rotate_encryption_key(db_session, old_key=OLD_KEY, dry_run=True)
+
+        assert totals.get("credential.credential_json", 0) >= 1
+
+        raw_after = _raw_credential_bytes(db_session, credential_id)
+        assert raw_after == original
+
+
+class TestRotateInternetSearchProvider:
+    """Test rotation against the real InternetSearchProvider table (EncryptedString).
+
+    Discovery is scoped to only the InternetSearchProvider model to avoid
+    mutating other tables in the test database.
+    """
+
+    @pytest.fixture(autouse=True)
+    def _limit_discovery(self) -> Generator[None, None, None]:
+        with patch(
+            f"{ROTATE_MODULE}._discover_encrypted_columns",
+            return_value=[
+                (InternetSearchProvider, "api_key", ["id"], False),
+            ],
+        ):
+            yield
+
+    @pytest.fixture()
+    def isp_id(
+        self, db_session: Session, tenant_context: None  # noqa: ARG002
+    ) -> Generator[int, None, None]:
+        """Insert an InternetSearchProvider row with raw encrypted bytes."""
+        encrypted = _encrypt_string("sk-secret-api-key", key=OLD_KEY)
+
+        result = db_session.execute(
+            text(
+                "INSERT INTO internet_search_provider "
+                "(name, provider_type, api_key, is_active) "
+                "VALUES (:name, :ptype, :api_key, false) "
+                "RETURNING id"
+            ),
+            {
+                "name": f"test-rotation-{id(self)}",
+                "ptype": "test",
+                "api_key": encrypted,
+            },
+        )
+        isp_id = result.scalar_one()
+        db_session.commit()
+
+        yield isp_id
+
+        db_session.execute(
+            text("DELETE FROM internet_search_provider WHERE id = :id"),
+            {"id": isp_id},
+        )
+        db_session.commit()
+
+    def test_rotates_api_key(self, db_session: Session, isp_id: int) -> None:
+        with (
+            patch(f"{ROTATE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+            patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+        ):
+            totals = rotate_encryption_key(db_session, old_key=OLD_KEY)
+
+        assert totals.get("internet_search_provider.api_key", 0) >= 1
+
+        raw = _raw_isp_bytes(db_session, isp_id)
+        assert raw is not None
+        assert _decrypt_bytes(raw, key=NEW_KEY) == "sk-secret-api-key"
+
+    def test_rotates_from_unencrypted(
+        self, db_session: Session, tenant_context: None  # noqa: ARG002
+    ) -> None:
+        """Test rotating data that was stored without any encryption key."""
+        result = db_session.execute(
+            text(
+                "INSERT INTO internet_search_provider "
+                "(name, provider_type, api_key, is_active) "
+                "VALUES (:name, :ptype, :api_key, false) "
+                "RETURNING id"
+            ),
+            {
+                "name": f"test-raw-{id(self)}",
+                "ptype": "test",
+                "api_key": b"raw-api-key",
+            },
+        )
+        isp_id = result.scalar_one()
+        db_session.commit()
+
+        try:
+            with (
+                patch(f"{ROTATE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+                patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", NEW_KEY),
+            ):
+                totals = rotate_encryption_key(db_session, old_key=None)
+
+            assert totals.get("internet_search_provider.api_key", 0) >= 1
+
+            raw = _raw_isp_bytes(db_session, isp_id)
+            assert raw is not None
+            assert _decrypt_bytes(raw, key=NEW_KEY) == "raw-api-key"
+        finally:
+            db_session.execute(
+                text("DELETE FROM internet_search_provider WHERE id = :id"),
+                {"id": isp_id},
+            )
+            db_session.commit()

backend/tests/external_dependency_unit/slack_bot/test_slack_bot_crud.py

@@ -0,0 +1,85 @@
+"""Tests that SlackBot CRUD operations return properly typed SensitiveValue fields.
+
+Regression test for the bug where insert_slack_bot/update_slack_bot returned
+objects with raw string tokens instead of SensitiveValue wrappers, causing
+'str object has no attribute get_value' errors in SlackBot.from_model().
+"""
+
+from uuid import uuid4
+
+from sqlalchemy.orm import Session
+
+from onyx.db.slack_bot import insert_slack_bot
+from onyx.db.slack_bot import update_slack_bot
+from onyx.server.manage.models import SlackBot
+from onyx.utils.sensitive import SensitiveValue
+
+
+def _unique(prefix: str) -> str:
+    return f"{prefix}-{uuid4().hex[:8]}"
+
+
+def test_insert_slack_bot_returns_sensitive_values(db_session: Session) -> None:
+    bot_token = _unique("xoxb-insert")
+    app_token = _unique("xapp-insert")
+    user_token = _unique("xoxp-insert")
+
+    slack_bot = insert_slack_bot(
+        db_session=db_session,
+        name=_unique("test-bot-insert"),
+        enabled=True,
+        bot_token=bot_token,
+        app_token=app_token,
+        user_token=user_token,
+    )
+
+    assert isinstance(slack_bot.bot_token, SensitiveValue)
+    assert isinstance(slack_bot.app_token, SensitiveValue)
+    assert isinstance(slack_bot.user_token, SensitiveValue)
+
+    assert slack_bot.bot_token.get_value(apply_mask=False) == bot_token
+    assert slack_bot.app_token.get_value(apply_mask=False) == app_token
+    assert slack_bot.user_token.get_value(apply_mask=False) == user_token
+
+    # Verify from_model works without error
+    pydantic_bot = SlackBot.from_model(slack_bot)
+    assert pydantic_bot.bot_token  # masked, but not empty
+    assert pydantic_bot.app_token
+
+
+def test_update_slack_bot_returns_sensitive_values(db_session: Session) -> None:
+    slack_bot = insert_slack_bot(
+        db_session=db_session,
+        name=_unique("test-bot-update"),
+        enabled=True,
+        bot_token=_unique("xoxb-update"),
+        app_token=_unique("xapp-update"),
+    )
+
+    new_bot_token = _unique("xoxb-update-new")
+    new_app_token = _unique("xapp-update-new")
+    new_user_token = _unique("xoxp-update-new")
+
+    updated = update_slack_bot(
+        db_session=db_session,
+        slack_bot_id=slack_bot.id,
+        name=_unique("test-bot-updated"),
+        enabled=False,
+        bot_token=new_bot_token,
+        app_token=new_app_token,
+        user_token=new_user_token,
+    )
+
+    assert isinstance(updated.bot_token, SensitiveValue)
+    assert isinstance(updated.app_token, SensitiveValue)
+    assert isinstance(updated.user_token, SensitiveValue)
+
+    assert updated.bot_token.get_value(apply_mask=False) == new_bot_token
+    assert updated.app_token.get_value(apply_mask=False) == new_app_token
+    assert updated.user_token.get_value(apply_mask=False) == new_user_token
+
+    # Verify from_model works without error
+    pydantic_bot = SlackBot.from_model(updated)
+    assert pydantic_bot.bot_token
+    assert pydantic_bot.app_token
+    assert pydantic_bot.user_token is not None

backend/tests/external_dependency_unit/tools/test_oauth_config_crud.py

@@ -148,8 +148,16 @@ class TestOAuthConfigCRUD:
         )
 
         # Secrets should be preserved
-        assert updated_config.client_id == original_client_id
-        assert updated_config.client_secret == original_client_secret
+        assert updated_config.client_id is not None
+        assert original_client_id is not None
+        assert updated_config.client_id.get_value(
+            apply_mask=False
+        ) == original_client_id.get_value(apply_mask=False)
+        assert updated_config.client_secret is not None
+        assert original_client_secret is not None
+        assert updated_config.client_secret.get_value(
+            apply_mask=False
+        ) == original_client_secret.get_value(apply_mask=False)
         # But name should be updated
         assert updated_config.name == new_name
 
@@ -173,9 +181,14 @@ class TestOAuthConfigCRUD:
         )
 
         # client_id should be cleared (empty string)
-        assert updated_config.client_id == ""
+        assert updated_config.client_id is not None
+        assert updated_config.client_id.get_value(apply_mask=False) == ""
         # client_secret should be preserved
-        assert updated_config.client_secret == original_client_secret
+        assert updated_config.client_secret is not None
+        assert original_client_secret is not None
+        assert updated_config.client_secret.get_value(
+            apply_mask=False
+        ) == original_client_secret.get_value(apply_mask=False)
 
     def test_update_oauth_config_clear_client_secret(self, db_session: Session) -> None:
         """Test clearing client_secret while preserving client_id"""
@@ -190,9 +203,14 @@ class TestOAuthConfigCRUD:
         )
 
         # client_secret should be cleared (empty string)
-        assert updated_config.client_secret == ""
+        assert updated_config.client_secret is not None
+        assert updated_config.client_secret.get_value(apply_mask=False) == ""
         # client_id should be preserved
-        assert updated_config.client_id == original_client_id
+        assert updated_config.client_id is not None
+        assert original_client_id is not None
+        assert updated_config.client_id.get_value(
+            apply_mask=False
+        ) == original_client_id.get_value(apply_mask=False)
 
     def test_update_oauth_config_clear_both_secrets(self, db_session: Session) -> None:
         """Test clearing both client_id and client_secret"""
@@ -207,8 +225,10 @@ class TestOAuthConfigCRUD:
         )
 
         # Both should be cleared (empty strings)
-        assert updated_config.client_id == ""
-        assert updated_config.client_secret == ""
+        assert updated_config.client_id is not None
+        assert updated_config.client_id.get_value(apply_mask=False) == ""
+        assert updated_config.client_secret is not None
+        assert updated_config.client_secret.get_value(apply_mask=False) == ""
 
     def test_update_oauth_config_authorization_url(self, db_session: Session) -> None:
         """Test updating authorization_url"""
@@ -275,7 +295,8 @@ class TestOAuthConfigCRUD:
         assert updated_config.token_url == new_token_url
         assert updated_config.scopes == new_scopes
         assert updated_config.additional_params == new_params
-        assert updated_config.client_id == new_client_id
+        assert updated_config.client_id is not None
+        assert updated_config.client_id.get_value(apply_mask=False) == new_client_id
 
     def test_delete_oauth_config(self, db_session: Session) -> None:
         """Test deleting an OAuth configuration"""
@@ -416,7 +437,8 @@ class TestOAuthUserTokenCRUD:
         assert user_token.id is not None
         assert user_token.oauth_config_id == oauth_config.id
         assert user_token.user_id == user.id
-        assert user_token.token_data == token_data
+        assert user_token.token_data is not None
+        assert user_token.token_data.get_value(apply_mask=False) == token_data
         assert user_token.created_at is not None
         assert user_token.updated_at is not None
 
@@ -446,8 +468,13 @@ class TestOAuthUserTokenCRUD:
 
         # Should be the same token record (updated, not inserted)
         assert updated_token.id == initial_token_id
-        assert updated_token.token_data == updated_token_data
-        assert updated_token.token_data != initial_token_data
+        assert updated_token.token_data is not None
+        assert (
+            updated_token.token_data.get_value(apply_mask=False) == updated_token_data
+        )
+        assert (
+            updated_token.token_data.get_value(apply_mask=False) != initial_token_data
+        )
 
     def test_get_user_oauth_token(self, db_session: Session) -> None:
         """Test retrieving a user's OAuth token"""
@@ -463,7 +490,8 @@ class TestOAuthUserTokenCRUD:
 
         assert retrieved_token is not None
         assert retrieved_token.id == created_token.id
-        assert retrieved_token.token_data == token_data
+        assert retrieved_token.token_data is not None
+        assert retrieved_token.token_data.get_value(apply_mask=False) == token_data
 
     def test_get_user_oauth_token_not_found(self, db_session: Session) -> None:
         """Test retrieving a non-existent user token returns None"""
@@ -519,7 +547,8 @@ class TestOAuthUserTokenCRUD:
         retrieved_token = get_user_oauth_token(oauth_config.id, user.id, db_session)
         assert retrieved_token is not None
         assert retrieved_token.id == updated_token.id
-        assert retrieved_token.token_data == token_data2
+        assert retrieved_token.token_data is not None
+        assert retrieved_token.token_data.get_value(apply_mask=False) == token_data2
 
     def test_cascade_delete_user_tokens_on_config_deletion(
         self, db_session: Session

backend/tests/external_dependency_unit/tools/test_oauth_token_manager.py

@@ -374,8 +374,14 @@ class TestOAuthTokenManagerCodeExchange:
         assert call_args[0][0] == oauth_config.token_url
         assert call_args[1]["data"]["grant_type"] == "authorization_code"
         assert call_args[1]["data"]["code"] == "auth_code_123"
-        assert call_args[1]["data"]["client_id"] == oauth_config.client_id
-        assert call_args[1]["data"]["client_secret"] == oauth_config.client_secret
+        assert oauth_config.client_id is not None
+        assert oauth_config.client_secret is not None
+        assert call_args[1]["data"]["client_id"] == oauth_config.client_id.get_value(
+            apply_mask=False
+        )
+        assert call_args[1]["data"][
+            "client_secret"
+        ] == oauth_config.client_secret.get_value(apply_mask=False)
         assert call_args[1]["data"]["redirect_uri"] == "https://example.com/callback"
 
     @patch("onyx.auth.oauth_token_manager.requests.post")

backend/tests/external_dependency_unit/tools/test_python_tool.py

@@ -950,6 +950,7 @@ from onyx.server.query_and_chat.streaming_models import Packet
 from onyx.server.query_and_chat.streaming_models import PythonToolDelta
 from onyx.server.query_and_chat.streaming_models import PythonToolStart
 from onyx.server.query_and_chat.streaming_models import SectionEnd
+from onyx.server.query_and_chat.streaming_models import ToolCallArgumentDelta
 from onyx.tools.tool_implementations.python.python_tool import PythonTool
 from tests.external_dependency_unit.answer.stream_test_builder import StreamTestBuilder
 from tests.external_dependency_unit.answer.stream_test_utils import create_chat_session
@@ -1294,9 +1295,18 @@ def test_code_interpreter_replay_packets_include_code_and_output(
             ).expect(
                 Packet(
                     placement=create_placement(0),
-                    obj=PythonToolStart(code=code),
+                    obj=ToolCallArgumentDelta(
+                        tool_type="python",
+                        argument_deltas={"code": code},
+                    ),
                 ),
                 forward=2,
+            ).expect(
+                Packet(
+                    placement=create_placement(0),
+                    obj=PythonToolStart(code=code),
+                ),
+                forward=False,
             ).expect(
                 Packet(
                     placement=create_placement(0),

backend/tests/integration/tests/discord_bot/test_discord_bot_db.py

@@ -64,7 +64,8 @@ class TestBotConfigAPI:
         db_session.commit()
 
         assert config is not None
-        assert config.bot_token == "test_token_123"
+        assert config.bot_token is not None
+        assert config.bot_token.get_value(apply_mask=False) == "test_token_123"
 
         # Cleanup
         delete_discord_bot_config(db_session)

backend/tests/unit/ee/onyx/utils/test_encryption.py

@@ -0,0 +1,165 @@
+"""Tests for EE AES-CBC encryption/decryption with explicit key support.
+
+With EE mode enabled (via conftest), fetch_versioned_implementation resolves
+to the EE implementations, so no patching of the MIT layer is needed.
+"""
+
+from unittest.mock import patch
+
+import pytest
+
+from ee.onyx.utils.encryption import _decrypt_bytes
+from ee.onyx.utils.encryption import _encrypt_string
+from ee.onyx.utils.encryption import _get_trimmed_key
+from ee.onyx.utils.encryption import decrypt_bytes_to_string
+from ee.onyx.utils.encryption import encrypt_string_to_bytes
+
+EE_MODULE = "ee.onyx.utils.encryption"
+
+# Keys must be exactly 16, 24, or 32 bytes for AES
+KEY_16 = "a" * 16
+KEY_16_ALT = "b" * 16
+KEY_24 = "d" * 24
+KEY_32 = "c" * 32
+
+
+@pytest.fixture(autouse=True)
+def _clear_key_cache() -> None:
+    _get_trimmed_key.cache_clear()
+
+
+class TestEncryptDecryptRoundTrip:
+    def test_roundtrip_with_env_key(self) -> None:
+        with patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", KEY_16):
+            encrypted = _encrypt_string("hello world")
+            assert encrypted != b"hello world"
+            assert _decrypt_bytes(encrypted) == "hello world"
+
+    def test_roundtrip_with_explicit_key(self) -> None:
+        encrypted = _encrypt_string("secret data", key=KEY_32)
+        assert encrypted != b"secret data"
+        assert _decrypt_bytes(encrypted, key=KEY_32) == "secret data"
+
+    def test_roundtrip_no_key(self) -> None:
+        """Without any key, data is raw-encoded (no encryption)."""
+        with patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", ""):
+            encrypted = _encrypt_string("plain text")
+            assert encrypted == b"plain text"
+            assert _decrypt_bytes(encrypted) == "plain text"
+
+    def test_explicit_key_overrides_env(self) -> None:
+        with patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", KEY_16):
+            encrypted = _encrypt_string("data", key=KEY_16_ALT)
+            with pytest.raises(ValueError):
+                _decrypt_bytes(encrypted, key=KEY_16)
+            assert _decrypt_bytes(encrypted, key=KEY_16_ALT) == "data"
+
+    def test_different_encryptions_produce_different_bytes(self) -> None:
+        """Each encryption uses a random IV, so results differ."""
+        a = _encrypt_string("same", key=KEY_16)
+        b = _encrypt_string("same", key=KEY_16)
+        assert a != b
+
+    def test_roundtrip_empty_string(self) -> None:
+        encrypted = _encrypt_string("", key=KEY_16)
+        assert encrypted != b""
+        assert _decrypt_bytes(encrypted, key=KEY_16) == ""
+
+    def test_roundtrip_unicode(self) -> None:
+        text = "日本語テスト 🔐 émojis"
+        encrypted = _encrypt_string(text, key=KEY_16)
+        assert _decrypt_bytes(encrypted, key=KEY_16) == text
+
+
+class TestDecryptFallbackBehavior:
+    def test_wrong_env_key_falls_back_to_raw_decode(self) -> None:
+        """Default key path: AES fails on non-AES data → fallback to raw decode."""
+        raw = "readable text".encode()
+        with patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", KEY_16):
+            assert _decrypt_bytes(raw) == "readable text"
+
+    def test_explicit_wrong_key_raises(self) -> None:
+        """Explicit key path: AES fails → raises, no fallback."""
+        encrypted = _encrypt_string("secret", key=KEY_16)
+        with pytest.raises(ValueError):
+            _decrypt_bytes(encrypted, key=KEY_16_ALT)
+
+    def test_explicit_none_key_with_no_env(self) -> None:
+        """key=None with empty env → raw decode."""
+        with patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", ""):
+            assert _decrypt_bytes(b"hello", key=None) == "hello"
+
+    def test_explicit_empty_string_key(self) -> None:
+        """key='' means no encryption."""
+        encrypted = _encrypt_string("test", key="")
+        assert encrypted == b"test"
+        assert _decrypt_bytes(encrypted, key="") == "test"
+
+
+class TestKeyValidation:
+    def test_key_too_short_raises(self) -> None:
+        with pytest.raises(RuntimeError, match="too short"):
+            _encrypt_string("data", key="short")
+
+    def test_16_byte_key(self) -> None:
+        encrypted = _encrypt_string("data", key=KEY_16)
+        assert _decrypt_bytes(encrypted, key=KEY_16) == "data"
+
+    def test_24_byte_key(self) -> None:
+        encrypted = _encrypt_string("data", key=KEY_24)
+        assert _decrypt_bytes(encrypted, key=KEY_24) == "data"
+
+    def test_32_byte_key(self) -> None:
+        encrypted = _encrypt_string("data", key=KEY_32)
+        assert _decrypt_bytes(encrypted, key=KEY_32) == "data"
+
+    def test_long_key_truncated_to_32(self) -> None:
+        """Keys longer than 32 bytes are truncated to 32."""
+        long_key = "e" * 64
+        encrypted = _encrypt_string("data", key=long_key)
+        assert _decrypt_bytes(encrypted, key=long_key) == "data"
+
+    def test_20_byte_key_trimmed_to_16(self) -> None:
+        """A 20-byte key is trimmed to the largest valid AES size that fits (16)."""
+        key_20 = "f" * 20
+        encrypted = _encrypt_string("data", key=key_20)
+        assert _decrypt_bytes(encrypted, key=key_20) == "data"
+
+        # Verify it was trimmed to 16 by checking that the first 16 bytes
+        # of the key can also decrypt it
+        key_16_same_prefix = "f" * 16
+        assert _decrypt_bytes(encrypted, key=key_16_same_prefix) == "data"
+
+    def test_25_byte_key_trimmed_to_24(self) -> None:
+        """A 25-byte key is trimmed to the largest valid AES size that fits (24)."""
+        key_25 = "g" * 25
+        encrypted = _encrypt_string("data", key=key_25)
+        assert _decrypt_bytes(encrypted, key=key_25) == "data"
+
+        key_24_same_prefix = "g" * 24
+        assert _decrypt_bytes(encrypted, key=key_24_same_prefix) == "data"
+
+    def test_30_byte_key_trimmed_to_24(self) -> None:
+        """A 30-byte key is trimmed to the largest valid AES size that fits (24)."""
+        key_30 = "h" * 30
+        encrypted = _encrypt_string("data", key=key_30)
+        assert _decrypt_bytes(encrypted, key=key_30) == "data"
+
+        key_24_same_prefix = "h" * 24
+        assert _decrypt_bytes(encrypted, key=key_24_same_prefix) == "data"
+
+
+class TestWrapperFunctions:
+    """Test encrypt_string_to_bytes / decrypt_bytes_to_string pass key through.
+
+    With EE mode enabled, the wrappers resolve to EE implementations automatically.
+    """
+
+    def test_wrapper_passes_key(self) -> None:
+        encrypted = encrypt_string_to_bytes("payload", key=KEY_16)
+        assert decrypt_bytes_to_string(encrypted, key=KEY_16) == "payload"
+
+    def test_wrapper_no_key_uses_env(self) -> None:
+        with patch(f"{EE_MODULE}.ENCRYPTION_KEY_SECRET", KEY_32):
+            encrypted = encrypt_string_to_bytes("payload")
+            assert decrypt_bytes_to_string(encrypted) == "payload"

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

@@ -0,0 +1,630 @@
+from typing import Any
+from unittest.mock import MagicMock
+from unittest.mock import patch
+
+from onyx.chat.tool_call_args_streaming import maybe_emit_argument_delta
+from onyx.server.query_and_chat.placement import Placement
+from onyx.server.query_and_chat.streaming_models import ToolCallArgumentDelta
+from onyx.utils.jsonriver import Parser
+
+
+def _make_tool_call_delta(
+    index: int = 0,
+    name: str | None = None,
+    arguments: str | None = None,
+    function_is_none: bool = False,
+) -> MagicMock:
+    """Create a mock tool_call_delta matching the LiteLLM streaming shape."""
+    delta = MagicMock()
+    delta.index = index
+    if function_is_none:
+        delta.function = None
+    else:
+        delta.function = MagicMock()
+        delta.function.name = name
+        delta.function.arguments = arguments
+    return delta
+
+
+def _make_placement() -> Placement:
+    return Placement(turn_index=0, tab_index=0)
+
+
+def _mock_tool_class(emit: bool = True) -> MagicMock:
+    cls = MagicMock()
+    cls.should_emit_argument_deltas.return_value = emit
+    return cls
+
+
+def _collect(
+    tc_map: dict[int, dict[str, Any]],
+    delta: MagicMock,
+    placement: Placement | None = None,
+    parsers: dict[int, Parser] | None = None,
+) -> list[Any]:
+    """Run maybe_emit_argument_delta and return the yielded packets."""
+    return list(
+        maybe_emit_argument_delta(
+            tc_map,
+            delta,
+            placement or _make_placement(),
+            parsers if parsers is not None else {},
+        )
+    )
+
+
+def _stream_fragments(
+    fragments: list[str],
+    tc_map: dict[int, dict[str, Any]],
+    placement: Placement | None = None,
+) -> list[str]:
+    """Feed fragments into maybe_emit_argument_delta one by one, returning
+    all emitted content values concatenated per-key as a flat list."""
+    pl = placement or _make_placement()
+    parsers: dict[int, Parser] = {}
+    emitted: list[str] = []
+    for frag in fragments:
+        tc_map[0]["arguments"] += frag
+        delta = _make_tool_call_delta(arguments=frag)
+        for packet in maybe_emit_argument_delta(tc_map, delta, pl, parsers=parsers):
+            obj = packet.obj
+            assert isinstance(obj, ToolCallArgumentDelta)
+            for value in obj.argument_deltas.values():
+                emitted.append(value)
+    return emitted
+
+
+class TestMaybeEmitArgumentDeltaGuards:
+    """Tests for conditions that cause no packet to be emitted."""
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_no_emission_when_tool_does_not_opt_in(
+        self, mock_get_tool: MagicMock
+    ) -> None:
+        """Tools that return False from should_emit_argument_deltas emit nothing."""
+        mock_get_tool.return_value = _mock_tool_class(emit=False)
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": '{"code": "x'}
+        }
+        assert _collect(tc_map, _make_tool_call_delta(arguments="x")) == []
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_no_emission_when_tool_class_unknown(
+        self, mock_get_tool: MagicMock
+    ) -> None:
+        mock_get_tool.return_value = None
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "unknown", "arguments": '{"code": "x'}
+        }
+        assert _collect(tc_map, _make_tool_call_delta(arguments="x")) == []
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_no_emission_when_no_argument_fragment(
+        self, mock_get_tool: MagicMock
+    ) -> None:
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": '{"code": "x'}
+        }
+        assert _collect(tc_map, _make_tool_call_delta(arguments=None)) == []
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_no_emission_when_key_value_not_started(
+        self, mock_get_tool: MagicMock
+    ) -> None:
+        """Key exists in JSON but its string value hasn't begun yet."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": '{"code":'}
+        }
+        assert _collect(tc_map, _make_tool_call_delta(arguments=":")) == []
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_no_emission_before_any_key(self, mock_get_tool: MagicMock) -> None:
+        """Only the opening brace has arrived — no key to stream yet."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": "{"}
+        }
+        assert _collect(tc_map, _make_tool_call_delta(arguments="{")) == []
+
+
+class TestMaybeEmitArgumentDeltaBasic:
+    """Tests for correct packet content and incremental emission."""
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_emits_packet_with_correct_fields(self, mock_get_tool: MagicMock) -> None:
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"code": "', "print(1)", '"}']
+
+        pl = _make_placement()
+        parsers: dict[int, Parser] = {}
+        all_packets = []
+        for frag in fragments:
+            tc_map[0]["arguments"] += frag
+            packets = _collect(
+                tc_map, _make_tool_call_delta(arguments=frag), pl, parsers
+            )
+            all_packets.extend(packets)
+
+        assert len(all_packets) >= 1
+        # Verify packet structure
+        obj = all_packets[0].obj
+        assert isinstance(obj, ToolCallArgumentDelta)
+        assert obj.tool_type == "python"
+        # All emitted content should reconstruct the value
+        full_code = ""
+        for p in all_packets:
+            assert isinstance(p.obj, ToolCallArgumentDelta)
+            if "code" in p.obj.argument_deltas:
+                full_code += p.obj.argument_deltas["code"]
+        assert full_code == "print(1)"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_emits_only_new_content_on_subsequent_call(
+        self, mock_get_tool: MagicMock
+    ) -> None:
+        """After a first emission, subsequent calls emit only the diff."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        parsers: dict[int, Parser] = {}
+        pl = _make_placement()
+
+        # First fragment opens the string
+        tc_map[0]["arguments"] = '{"code": "abc'
+        packets_1 = _collect(
+            tc_map, _make_tool_call_delta(arguments='{"code": "abc'), pl, parsers
+        )
+        code_1 = ""
+        for p in packets_1:
+            assert isinstance(p.obj, ToolCallArgumentDelta)
+            code_1 += p.obj.argument_deltas.get("code", "")
+        assert code_1 == "abc"
+
+        # Second fragment appends more
+        tc_map[0]["arguments"] = '{"code": "abcdef'
+        packets_2 = _collect(
+            tc_map, _make_tool_call_delta(arguments="def"), pl, parsers
+        )
+        code_2 = ""
+        for p in packets_2:
+            assert isinstance(p.obj, ToolCallArgumentDelta)
+            code_2 += p.obj.argument_deltas.get("code", "")
+        assert code_2 == "def"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_handles_multiple_keys_sequentially(self, mock_get_tool: MagicMock) -> None:
+        """When a second key starts, emissions switch to that key."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = [
+            '{"code": "x',
+            '", "output": "hello',
+            '"}',
+        ]
+
+        emitted = _stream_fragments(fragments, tc_map)
+        full = "".join(emitted)
+        assert "x" in full
+        assert "hello" in full
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_delta_spans_key_boundary(self, mock_get_tool: MagicMock) -> None:
+        """A single delta contains the end of one value and the start of the next key."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = [
+            '{"code": "x',
+            'y", "lang": "py',
+            '"}',
+        ]
+
+        emitted = _stream_fragments(fragments, tc_map)
+        full = "".join(emitted)
+        assert "xy" in full
+        assert "py" in full
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_empty_value_emits_nothing(self, mock_get_tool: MagicMock) -> None:
+        """An empty string value has nothing to emit."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        # Opening quote just arrived, value is empty
+        tc_map[0]["arguments"] = '{"code": "'
+        packets = _collect(tc_map, _make_tool_call_delta(arguments='{"code": "'))
+        # No string content yet, so either no packet or empty deltas
+        for p in packets:
+            assert isinstance(p.obj, ToolCallArgumentDelta)
+            assert p.obj.argument_deltas.get("code", "") == ""
+
+
+class TestMaybeEmitArgumentDeltaDecoding:
+    """Tests verifying that JSON escape sequences are properly decoded."""
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_decodes_newlines(self, mock_get_tool: MagicMock) -> None:
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"code": "line1\\nline2"}']
+
+        emitted = _stream_fragments(fragments, tc_map)
+        assert "".join(emitted) == "line1\nline2"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_decodes_tabs(self, mock_get_tool: MagicMock) -> None:
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"code": "\\tindented"}']
+
+        emitted = _stream_fragments(fragments, tc_map)
+        assert "".join(emitted) == "\tindented"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_decodes_escaped_quotes(self, mock_get_tool: MagicMock) -> None:
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"code": "say \\"hi\\""}']
+
+        emitted = _stream_fragments(fragments, tc_map)
+        assert "".join(emitted) == 'say "hi"'
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_decodes_escaped_backslashes(self, mock_get_tool: MagicMock) -> None:
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"code": "path\\\\dir"}']
+
+        emitted = _stream_fragments(fragments, tc_map)
+        assert "".join(emitted) == "path\\dir"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_decodes_unicode_escape(self, mock_get_tool: MagicMock) -> None:
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"code": "\\u0041"}']
+
+        emitted = _stream_fragments(fragments, tc_map)
+        assert "".join(emitted) == "A"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_incomplete_escape_at_end_decoded_on_next_chunk(
+        self, mock_get_tool: MagicMock
+    ) -> None:
+        """A trailing backslash (incomplete escape) is completed in the next chunk."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"code": "hello\\', 'n"}']
+
+        emitted = _stream_fragments(fragments, tc_map)
+        assert "".join(emitted) == "hello\n"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_incomplete_unicode_escape_completed_on_next_chunk(
+        self, mock_get_tool: MagicMock
+    ) -> None:
+        """A partial \\uXX sequence is completed in the next chunk."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"code": "hello\\u00', '41"}']
+
+        emitted = _stream_fragments(fragments, tc_map)
+        assert "".join(emitted) == "helloA"
+
+
+class TestArgumentDeltaStreamingE2E:
+    """Simulates realistic sequences of LLM argument deltas to verify
+    the full pipeline produces correct decoded output."""
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_realistic_python_code_streaming(self, mock_get_tool: MagicMock) -> None:
+        """Streams: {"code": "print('hello')\\nprint('world')"}"""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = [
+            '{"',
+            "code",
+            '": "',
+            "print(",
+            "'hello')",
+            "\\n",
+            "print(",
+            "'world')",
+            '"}',
+        ]
+
+        full = "".join(_stream_fragments(fragments, tc_map))
+        assert full == "print('hello')\nprint('world')"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_streaming_with_tabs_and_newlines(self, mock_get_tool: MagicMock) -> None:
+        """Streams code with tabs and newlines."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = [
+            '{"code": "',
+            "if True:",
+            "\\n",
+            "\\t",
+            "pass",
+            '"}',
+        ]
+
+        full = "".join(_stream_fragments(fragments, tc_map))
+        assert full == "if True:\n\tpass"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_split_escape_sequence(self, mock_get_tool: MagicMock) -> None:
+        """An escape sequence split across two fragments (backslash in one,
+        'n' in the next) should still decode correctly."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = [
+            '{"code": "hello',
+            "\\",
+            "n",
+            'world"}',
+        ]
+
+        full = "".join(_stream_fragments(fragments, tc_map))
+        assert full == "hello\nworld"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_multiple_newlines_and_indentation(self, mock_get_tool: MagicMock) -> None:
+        """Streams a multi-line function with multiple escape sequences."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = [
+            '{"code": "',
+            "def foo():",
+            "\\n",
+            "\\t",
+            "x = 1",
+            "\\n",
+            "\\t",
+            "return x",
+            '"}',
+        ]
+
+        full = "".join(_stream_fragments(fragments, tc_map))
+        assert full == "def foo():\n\tx = 1\n\treturn x"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_two_keys_streamed_sequentially(self, mock_get_tool: MagicMock) -> None:
+        """Streams code first, then a second key (language) — both decoded."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = [
+            '{"code": "',
+            "x = 1",
+            '", "language": "',
+            "python",
+            '"}',
+        ]
+
+        emitted = _stream_fragments(fragments, tc_map)
+        # Should have emissions for both keys
+        full = "".join(emitted)
+        assert "x = 1" in full
+        assert "python" in full
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_code_containing_dict_literal(self, mock_get_tool: MagicMock) -> None:
+        """Python code like `x = {"key": "val"}` contains JSON-like patterns.
+        The escaped quotes inside the *outer* JSON value should prevent the
+        inner `"key":` from being mistaken for a top-level JSON key."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        # The LLM sends: {"code": "x = {\"key\": \"val\"}"}
+        # The inner quotes are escaped as \" in the JSON value.
+        fragments = [
+            '{"code": "',
+            "x = {",
+            '\\"key\\"',
+            ": ",
+            '\\"val\\"',
+            "}",
+            '"}',
+        ]
+
+        full = "".join(_stream_fragments(fragments, tc_map))
+        assert full == 'x = {"key": "val"}'
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_code_with_colon_in_value(self, mock_get_tool: MagicMock) -> None:
+        """Colons inside the string value should not confuse key detection."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = [
+            '{"code": "',
+            "url = ",
+            '\\"https://example.com\\"',
+            '"}',
+        ]
+
+        full = "".join(_stream_fragments(fragments, tc_map))
+        assert full == 'url = "https://example.com"'
+
+
+class TestMaybeEmitArgumentDeltaEdgeCases:
+    """Edge cases not covered by the standard test classes."""
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_no_emission_when_function_is_none(self, mock_get_tool: MagicMock) -> None:
+        """Some delta chunks have function=None (e.g. role-only deltas)."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": '{"code": "x'}
+        }
+        delta = _make_tool_call_delta(arguments=None, function_is_none=True)
+        assert _collect(tc_map, delta) == []
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_multiple_concurrent_tool_calls(self, mock_get_tool: MagicMock) -> None:
+        """Two tool calls streaming at different indices in parallel."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""},
+            1: {"id": "tc_2", "name": "python", "arguments": ""},
+        }
+
+        parsers: dict[int, Parser] = {}
+        pl = _make_placement()
+
+        # Feed full JSON to index 0
+        tc_map[0]["arguments"] = '{"code": "aaa"}'
+        packets_0 = _collect(
+            tc_map,
+            _make_tool_call_delta(index=0, arguments='{"code": "aaa"}'),
+            pl,
+            parsers,
+        )
+        code_0 = ""
+        for p in packets_0:
+            assert isinstance(p.obj, ToolCallArgumentDelta)
+            code_0 += p.obj.argument_deltas.get("code", "")
+        assert code_0 == "aaa"
+
+        # Feed full JSON to index 1
+        tc_map[1]["arguments"] = '{"code": "bbb"}'
+        packets_1 = _collect(
+            tc_map,
+            _make_tool_call_delta(index=1, arguments='{"code": "bbb"}'),
+            pl,
+            parsers,
+        )
+        code_1 = ""
+        for p in packets_1:
+            assert isinstance(p.obj, ToolCallArgumentDelta)
+            code_1 += p.obj.argument_deltas.get("code", "")
+        assert code_1 == "bbb"
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_delta_with_four_arguments(self, mock_get_tool: MagicMock) -> None:
+        """A single delta contains four complete key-value pairs."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        full = '{"a": "one", "b": "two", "c": "three", "d": "four"}'
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        tc_map[0]["arguments"] = full
+        parsers: dict[int, Parser] = {}
+        packets = _collect(
+            tc_map, _make_tool_call_delta(arguments=full), parsers=parsers
+        )
+
+        # Collect all argument deltas across packets
+        all_deltas: dict[str, str] = {}
+        for p in packets:
+            assert isinstance(p.obj, ToolCallArgumentDelta)
+            for k, v in p.obj.argument_deltas.items():
+                all_deltas[k] = all_deltas.get(k, "") + v
+
+        assert all_deltas == {
+            "a": "one",
+            "b": "two",
+            "c": "three",
+            "d": "four",
+        }
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_delta_on_second_arg_after_first_complete(
+        self, mock_get_tool: MagicMock
+    ) -> None:
+        """First argument is fully complete; delta only adds to the second."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+
+        fragments = [
+            '{"code": "print(1)", "lang": "py',
+            '"}',
+        ]
+
+        emitted = _stream_fragments(fragments, tc_map)
+        full = "".join(emitted)
+        assert "print(1)" in full
+        assert "py" in full
+
+    @patch("onyx.chat.tool_call_args_streaming._get_tool_class")
+    def test_non_string_values_skipped(self, mock_get_tool: MagicMock) -> None:
+        """Non-string values (numbers, booleans, null) are skipped — they are
+        available in the final tool-call kickoff packet. String arguments
+        following them are still emitted."""
+        mock_get_tool.return_value = _mock_tool_class()
+
+        tc_map: dict[int, dict[str, Any]] = {
+            0: {"id": "tc_1", "name": "python", "arguments": ""}
+        }
+        fragments = ['{"timeout": 30, "code": "hello"}']
+
+        emitted = _stream_fragments(fragments, tc_map)
+        full = "".join(emitted)
+        assert full == "hello"

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

@@ -0,0 +1,188 @@
+from io import BytesIO
+from unittest.mock import MagicMock
+
+import pytest
+from fastapi import UploadFile
+
+from onyx.server.features.projects import projects_file_utils as utils
+
+
+class _Tokenizer:
+    def encode(self, text: str) -> list[int]:
+        return [1] * len(text)
+
+
+class _NonSeekableFile(BytesIO):
+    def tell(self) -> int:
+        raise OSError("tell not supported")
+
+    def seek(self, *_args: object, **_kwargs: object) -> int:
+        raise OSError("seek not supported")
+
+
+def _make_upload(filename: str, size: int, content: bytes | None = None) -> UploadFile:
+    payload = content if content is not None else (b"x" * size)
+    return UploadFile(filename=filename, file=BytesIO(payload), size=size)
+
+
+def _make_upload_no_size(filename: str, content: bytes) -> UploadFile:
+    return UploadFile(filename=filename, file=BytesIO(content), size=None)
+
+
+def _patch_common_dependencies(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setattr(utils, "fetch_default_llm_model", lambda _db: None)
+    monkeypatch.setattr(utils, "get_tokenizer", lambda **_kwargs: _Tokenizer())
+    monkeypatch.setattr(utils, "is_file_password_protected", lambda **_kwargs: False)
+
+
+def test_get_upload_size_bytes_falls_back_to_stream_size() -> None:
+    upload = UploadFile(filename="example.txt", file=BytesIO(b"abcdef"), size=None)
+    upload.file.seek(2)
+
+    size = utils.get_upload_size_bytes(upload)
+
+    assert size == 6
+    assert upload.file.tell() == 2
+
+
+def test_get_upload_size_bytes_logs_warning_when_stream_size_unavailable(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    upload = UploadFile(filename="non_seekable.txt", file=_NonSeekableFile(), size=None)
+
+    caplog.set_level("WARNING")
+    size = utils.get_upload_size_bytes(upload)
+
+    assert size is None
+    assert "Could not determine upload size via stream seek" in caplog.text
+    assert "non_seekable.txt" in caplog.text
+
+
+def test_is_upload_too_large_logs_warning_when_size_unknown(
+    monkeypatch: pytest.MonkeyPatch,
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    upload = _make_upload("size_unknown.txt", size=1)
+    monkeypatch.setattr(utils, "get_upload_size_bytes", lambda _upload: None)
+
+    caplog.set_level("WARNING")
+    is_too_large = utils.is_upload_too_large(upload, max_bytes=100)
+
+    assert is_too_large is False
+    assert "Could not determine upload size; skipping size-limit check" in caplog.text
+    assert "size_unknown.txt" in caplog.text
+
+
+def test_categorize_uploaded_files_accepts_size_under_limit(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_common_dependencies(monkeypatch)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_BYTES", 100)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_MB", 1)
+    monkeypatch.setattr(utils, "estimate_image_tokens_for_upload", lambda _upload: 10)
+
+    upload = _make_upload("small.png", size=99)
+    result = utils.categorize_uploaded_files([upload], MagicMock())
+
+    assert len(result.acceptable) == 1
+    assert len(result.rejected) == 0
+
+
+def test_categorize_uploaded_files_uses_seek_fallback_when_upload_size_missing(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_common_dependencies(monkeypatch)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_BYTES", 100)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_MB", 1)
+    monkeypatch.setattr(utils, "estimate_image_tokens_for_upload", lambda _upload: 10)
+
+    upload = _make_upload_no_size("small.png", content=b"x" * 99)
+    result = utils.categorize_uploaded_files([upload], MagicMock())
+
+    assert len(result.acceptable) == 1
+    assert len(result.rejected) == 0
+
+
+def test_categorize_uploaded_files_accepts_size_at_limit(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_common_dependencies(monkeypatch)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_BYTES", 100)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_MB", 1)
+    monkeypatch.setattr(utils, "estimate_image_tokens_for_upload", lambda _upload: 10)
+
+    upload = _make_upload("edge.png", size=100)
+    result = utils.categorize_uploaded_files([upload], MagicMock())
+
+    assert len(result.acceptable) == 1
+    assert len(result.rejected) == 0
+
+
+def test_categorize_uploaded_files_rejects_size_over_limit_with_reason(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_common_dependencies(monkeypatch)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_BYTES", 100)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_MB", 1)
+    monkeypatch.setattr(utils, "estimate_image_tokens_for_upload", lambda _upload: 10)
+
+    upload = _make_upload("large.png", size=101)
+    result = utils.categorize_uploaded_files([upload], MagicMock())
+
+    assert len(result.acceptable) == 0
+    assert len(result.rejected) == 1
+    assert result.rejected[0].reason == "Exceeds 1 MB file size limit"
+
+
+def test_categorize_uploaded_files_mixed_batch_keeps_valid_and_rejects_oversized(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_common_dependencies(monkeypatch)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_BYTES", 100)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_MB", 1)
+    monkeypatch.setattr(utils, "estimate_image_tokens_for_upload", lambda _upload: 10)
+
+    small = _make_upload("small.png", size=50)
+    large = _make_upload("large.png", size=101)
+
+    result = utils.categorize_uploaded_files([small, large], MagicMock())
+
+    assert [file.filename for file in result.acceptable] == ["small.png"]
+    assert len(result.rejected) == 1
+    assert result.rejected[0].filename == "large.png"
+    assert result.rejected[0].reason == "Exceeds 1 MB file size limit"
+
+
+def test_categorize_uploaded_files_enforces_size_limit_even_when_threshold_is_skipped(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_common_dependencies(monkeypatch)
+    monkeypatch.setattr(utils, "SKIP_USERFILE_THRESHOLD", True)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_BYTES", 100)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_MB", 1)
+
+    upload = _make_upload("oversized.pdf", size=101)
+    result = utils.categorize_uploaded_files([upload], MagicMock())
+
+    assert len(result.acceptable) == 0
+    assert len(result.rejected) == 1
+    assert result.rejected[0].reason == "Exceeds 1 MB file size limit"
+
+
+def test_categorize_uploaded_files_checks_size_before_text_extraction(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    _patch_common_dependencies(monkeypatch)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_BYTES", 100)
+    monkeypatch.setattr(utils, "USER_FILE_MAX_UPLOAD_SIZE_MB", 1)
+
+    extract_mock = MagicMock(return_value="this should not run")
+    monkeypatch.setattr(utils, "extract_file_text", extract_mock)
+
+    oversized_doc = _make_upload("oversized.pdf", size=101)
+    result = utils.categorize_uploaded_files([oversized_doc], MagicMock())
+
+    extract_mock.assert_not_called()
+    assert len(result.acceptable) == 0
+    assert len(result.rejected) == 1
+    assert result.rejected[0].reason == "Exceeds 1 MB file size limit"

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

@@ -0,0 +1,32 @@
+import pytest
+
+from onyx.key_value_store.interface import KvKeyNotFoundError
+from onyx.server.settings import store as settings_store
+
+
+class _FakeKvStore:
+    def load(self, _key: str) -> dict:
+        raise KvKeyNotFoundError()
+
+
+class _FakeCache:
+    def __init__(self) -> None:
+        self._vals: dict[str, bytes] = {}
+
+    def get(self, key: str) -> bytes | None:
+        return self._vals.get(key)
+
+    def set(self, key: str, value: str, ex: int | None = None) -> None:  # noqa: ARG002
+        self._vals[key] = value.encode("utf-8")
+
+
+def test_load_settings_includes_user_file_max_upload_size_mb(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(settings_store, "get_kv_store", lambda: _FakeKvStore())
+    monkeypatch.setattr(settings_store, "get_cache_backend", lambda: _FakeCache())
+    monkeypatch.setattr(settings_store, "USER_FILE_MAX_UPLOAD_SIZE_MB", 77)
+
+    settings = settings_store.load_settings()
+
+    assert settings.user_file_max_upload_size_mb == 77

backend/tests/unit/onyx/utils/test_sensitive.py

@@ -147,15 +147,18 @@ class TestSensitiveValueString:
         )
         assert sensitive1 != sensitive2
 
-    def test_equality_with_non_sensitive_raises(self) -> None:
-        """Test that comparing with non-SensitiveValue raises error."""
+    def test_equality_with_non_sensitive_returns_not_equal(self) -> None:
+        """Test that comparing with non-SensitiveValue is always not-equal.
+
+        Returns NotImplemented so Python falls back to identity comparison.
+        This is required for compatibility with SQLAlchemy's attribute tracking.
+        """
         sensitive = SensitiveValue(
             encrypted_bytes=_encrypt_string("secret"),
             decrypt_fn=_decrypt_string,
             is_json=False,
         )
-        with pytest.raises(SensitiveAccessError):
-            _ = sensitive == "secret"
+        assert not (sensitive == "secret")
 
 
 class TestSensitiveValueJson:

deployment/helm/charts/onyx/values.yaml

@@ -1261,3 +1261,5 @@ configMap:
   SKIP_USERFILE_THRESHOLD: ""
   # For multi-tenant: comma-separated list of tenant IDs to skip threshold
   SKIP_USERFILE_THRESHOLD_TENANT_IDS: ""
+  # Maximum user upload file size in MB for chat/projects uploads
+  USER_FILE_MAX_UPLOAD_SIZE_MB: ""

web/jest.config.js

@@ -156,6 +156,7 @@ module.exports = {
         "**/src/app/**/*.test.tsx",
         "**/src/components/**/*.test.tsx",
         "**/src/lib/**/*.test.tsx",
+        "**/src/providers/**/*.test.tsx",
         "**/src/refresh-components/**/*.test.tsx",
         "**/src/hooks/**/*.test.tsx",
         "**/src/sections/**/*.test.tsx",

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

@@ -39,7 +39,7 @@ type ButtonProps = InteractiveStatelessProps &
     /** Tooltip text shown on hover. */
     tooltip?: string;
 
-    /** Width preset. `"auto"` shrink-wraps, `"full"` stretches to parent width. */
+    /** Width preset. `"fit"` shrink-wraps, `"full"` stretches to parent width. */
     width?: WidthVariant;
 
     /** Which side the tooltip appears on. */

web/lib/opal/src/components/buttons/line-item-button/README.md

@@ -0,0 +1,83 @@
+# LineItemButton
+
+**Import:** `import { LineItemButton, type LineItemButtonProps } from "@opal/components";`
+
+A composite component that wraps `Interactive.Stateful > Interactive.Container > ContentAction` into a single API. Use it for selectable list rows such as model pickers, menu items, or any row that acts like a button.
+
+## Architecture
+
+```
+Interactive.Stateful         <- selectVariant, state, interaction, onClick, href, ref
+  └─ Interactive.Container   <- type, width, roundingVariant
+       └─ ContentAction      <- withInteractive, paddingVariant="lg"
+            ├─ Content       <- icon, title, description, sizePreset, variant, ...
+            └─ rightChildren
+```
+
+`paddingVariant` is hardcoded to `"lg"` and `withInteractive` is always `true`. These are not exposed as props.
+
+## Props
+
+### Interactive surface
+
+| Prop | Type | Default | Description |
+|------|------|---------|-------------|
+| `selectVariant` | `"select-light" \| "select-heavy"` | `"select-light"` | Interactive select variant |
+| `state` | `InteractiveStatefulState` | `"empty"` | Value state (`"empty"`, `"filled"`, `"selected"`) |
+| `interaction` | `InteractiveStatefulInteraction` | `"rest"` | JS-controlled interaction state override |
+| `onClick` | `MouseEventHandler<HTMLElement>` | — | Click handler |
+| `href` | `string` | — | Renders an anchor instead of a div |
+| `target` | `string` | — | Anchor target (e.g. `"_blank"`) |
+| `group` | `string` | — | Interactive group key |
+| `ref` | `React.Ref<HTMLElement>` | — | Forwarded ref |
+
+### Sizing
+
+| Prop | Type | Default | Description |
+|------|------|---------|-------------|
+| `roundingVariant` | `InteractiveContainerRoundingVariant` | `"default"` | Corner rounding preset (height is content-driven) |
+| `width` | `WidthVariant` | `"full"` | Container width |
+| `type` | `"submit" \| "button" \| "reset"` | `"button"` | HTML button type |
+| `tooltip` | `string` | — | Tooltip text shown on hover |
+| `tooltipSide` | `TooltipSide` | `"top"` | Tooltip side |
+
+### Content (pass-through to ContentAction)
+
+| Prop | Type | Default | Description |
+|------|------|---------|-------------|
+| `title` | `string` | **(required)** | Row label |
+| `icon` | `IconFunctionComponent` | — | Left icon |
+| `description` | `string` | — | Description below the title |
+| `sizePreset` | `SizePreset` | `"headline"` | Content size preset |
+| `variant` | `ContentVariant` | `"heading"` | Content layout variant |
+| `rightChildren` | `ReactNode` | — | Content after the label (e.g. action button) |
+
+All other `ContentAction` / `Content` props (`editable`, `onTitleChange`, `optional`, `auxIcon`, `tag`, etc.) are also passed through. Note: `withInteractive` is always `true` inside `LineItemButton` and cannot be overridden.
+
+## Usage
+
+```tsx
+import { LineItemButton } from "@opal/components";
+
+// Simple selectable row
+<LineItemButton
+  selectVariant="select-heavy"
+  state={isSelected ? "selected" : "empty"}
+  roundingVariant="compact"
+  onClick={handleClick}
+  title="gpt-4o"
+  sizePreset="main-ui"
+  variant="section"
+/>
+
+// With right-side action
+<LineItemButton
+  selectVariant="select-heavy"
+  state={isSelected ? "selected" : "empty"}
+  onClick={handleClick}
+  title="claude-opus-4"
+  sizePreset="main-ui"
+  variant="section"
+  rightChildren={<Tag title="Default" color="blue" />}
+/>
+```

web/lib/opal/src/components/buttons/line-item-button/components.tsx

@@ -0,0 +1,137 @@
+import "@opal/components/tooltip.css";
+import {
+  Interactive,
+  type InteractiveStatefulState,
+  type InteractiveStatefulInteraction,
+  type InteractiveStatefulProps,
+  InteractiveContainerRoundingVariant,
+} from "@opal/core";
+import { type WidthVariant } from "@opal/shared";
+import type { TooltipSide } from "@opal/components";
+import type { DistributiveOmit } from "@opal/types";
+import type { ContentActionProps } from "@opal/layouts/content-action/components";
+import { ContentAction } from "@opal/layouts";
+import * as TooltipPrimitive from "@radix-ui/react-tooltip";
+
+// ---------------------------------------------------------------------------
+// Types
+// ---------------------------------------------------------------------------
+
+type ContentPassthroughProps = DistributiveOmit<
+  ContentActionProps,
+  "paddingVariant" | "widthVariant" | "ref" | "withInteractive"
+>;
+
+type LineItemButtonOwnProps = {
+  /** Interactive select variant. @default "select-light" */
+  selectVariant?: "select-light" | "select-heavy";
+
+  /** Value state. @default "empty" */
+  state?: InteractiveStatefulState;
+
+  /** JS-controllable interaction state override. @default "rest" */
+  interaction?: InteractiveStatefulInteraction;
+
+  /** Click handler. */
+  onClick?: InteractiveStatefulProps["onClick"];
+
+  /** When provided, renders an anchor instead of a div. */
+  href?: string;
+
+  /** Anchor target (e.g. "_blank"). */
+  target?: string;
+
+  /** Interactive group key. */
+  group?: string;
+
+  /** Forwarded ref. */
+  ref?: React.Ref<HTMLElement>;
+
+  /** Corner rounding preset (height is always content-driven). @default "default" */
+  roundingVariant?: InteractiveContainerRoundingVariant;
+
+  /** Container width. @default "full" */
+  width?: WidthVariant;
+
+  /** HTML button type. @default "button" */
+  type?: "submit" | "button" | "reset";
+
+  /** Tooltip text shown on hover. */
+  tooltip?: string;
+
+  /** Which side the tooltip appears on. @default "top" */
+  tooltipSide?: TooltipSide;
+};
+
+type LineItemButtonProps = ContentPassthroughProps & LineItemButtonOwnProps;
+
+// ---------------------------------------------------------------------------
+// LineItemButton
+// ---------------------------------------------------------------------------
+
+function LineItemButton({
+  // Interactive surface
+  selectVariant = "select-light",
+  state,
+  interaction,
+  onClick,
+  href,
+  target,
+  group,
+  ref,
+
+  // Sizing
+  roundingVariant = "default",
+  width = "full",
+  type = "button",
+  tooltip,
+  tooltipSide = "top",
+
+  // ContentAction pass-through
+  ...contentActionProps
+}: LineItemButtonProps) {
+  const item = (
+    <Interactive.Stateful
+      variant={selectVariant}
+      state={state}
+      interaction={interaction}
+      onClick={onClick}
+      href={href}
+      target={target}
+      group={group}
+      ref={ref}
+    >
+      <Interactive.Container
+        type={type}
+        widthVariant={width}
+        heightVariant="lg"
+        roundingVariant={roundingVariant}
+      >
+        <ContentAction
+          {...(contentActionProps as ContentActionProps)}
+          withInteractive
+          paddingVariant="fit"
+        />
+      </Interactive.Container>
+    </Interactive.Stateful>
+  );
+
+  if (!tooltip) return item;
+
+  return (
+    <TooltipPrimitive.Root>
+      <TooltipPrimitive.Trigger asChild>{item}</TooltipPrimitive.Trigger>
+      <TooltipPrimitive.Portal>
+        <TooltipPrimitive.Content
+          className="opal-tooltip"
+          side={tooltipSide}
+          sideOffset={4}
+        >
+          {tooltip}
+        </TooltipPrimitive.Content>
+      </TooltipPrimitive.Portal>
+    </TooltipPrimitive.Root>
+  );
+}
+
+export { LineItemButton, type LineItemButtonProps };

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

@@ -56,7 +56,7 @@ type SelectButtonProps = InteractiveStatefulProps &
     /** Tooltip text shown on hover. */
     tooltip?: string;
 
-    /** Width preset. `"auto"` shrink-wraps, `"full"` stretches to parent width. */
+    /** Width preset. `"fit"` shrink-wraps, `"full"` stretches to parent width. */
     width?: WidthVariant;
 
     /** Which side the tooltip appears on. */

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

@@ -19,6 +19,12 @@ export {
   type OpenButtonProps,
 } from "@opal/components/buttons/open-button/components";
 
+/* LineItemButton */
+export {
+  LineItemButton,
+  type LineItemButtonProps,
+} from "@opal/components/buttons/line-item-button/components";
+
 /* Tag */
 export {
   Tag,

web/lib/opal/src/core/animations/components.tsx

@@ -2,6 +2,7 @@ import "@opal/core/animations/styles.css";
 import React, { createContext, useContext, useState, useCallback } from "react";
 import { cn } from "@opal/utils";
 import type { WithoutStyles } from "@opal/types";
+import { widthVariants, type WidthVariant } from "@opal/shared";
 
 // ---------------------------------------------------------------------------
 // Context-per-group registry
@@ -38,6 +39,10 @@ interface HoverableRootProps
   extends WithoutStyles<React.HTMLAttributes<HTMLDivElement>> {
   children: React.ReactNode;
   group: string;
+  /** Width preset. @default "auto" */
+  widthVariant?: WidthVariant;
+  /** Ref forwarded to the root `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 type HoverableItemVariant = "opacity-on-hover";
@@ -47,6 +52,8 @@ interface HoverableItemProps
   children: React.ReactNode;
   group?: string;
   variant?: HoverableItemVariant;
+  /** Ref forwarded to the item `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -77,6 +84,8 @@ interface HoverableItemProps
 function HoverableRoot({
   group,
   children,
+  widthVariant = "auto",
+  ref,
   onMouseEnter: consumerMouseEnter,
   onMouseLeave: consumerMouseLeave,
   ...props
@@ -103,7 +112,13 @@ function HoverableRoot({
 
   return (
     <GroupContext.Provider value={hovered}>
-      <div {...props} onMouseEnter={onMouseEnter} onMouseLeave={onMouseLeave}>
+      <div
+        {...props}
+        ref={ref}
+        className={cn(widthVariants[widthVariant])}
+        onMouseEnter={onMouseEnter}
+        onMouseLeave={onMouseLeave}
+      >
         {children}
       </div>
     </GroupContext.Provider>
@@ -147,6 +162,7 @@ function HoverableItem({
   group,
   variant = "opacity-on-hover",
   children,
+  ref,
   ...props
 }: HoverableItemProps) {
   const contextValue = useContext(
@@ -165,6 +181,7 @@ function HoverableItem({
   return (
     <div
       {...props}
+      ref={ref}
       className={cn("hoverable-item")}
       data-hoverable-variant={variant}
       data-hoverable-active={

web/lib/opal/src/core/interactive/container/README.md

@@ -10,7 +10,7 @@ Structural container shared by both `Interactive.Stateless` and `Interactive.Sta
 |------|------|---------|-------------|
 | `heightVariant` | `SizeVariant` | `"lg"` | Height preset (`2xs`–`lg`, `fit`) |
 | `roundingVariant` | `"default" \| "compact" \| "mini"` | `"default"` | Border-radius preset |
-| `widthVariant` | `WidthVariant` | — | Width preset (`auto`, `full`) |
+| `widthVariant` | `WidthVariant` | — | Width preset (`"auto"`, `"fit"`, `"full"`) |
 | `border` | `boolean` | `false` | Renders a 1px border |
 | `type` | `"submit" \| "button" \| "reset"` | — | When set, renders a `<button>` element |
 

web/lib/opal/src/core/interactive/container/components.tsx

@@ -78,7 +78,7 @@ interface InteractiveContainerProps
   /**
    * Width preset controlling the container's horizontal size.
    *
-   * @default "auto"
+   * @default "fit"
    */
   widthVariant?: WidthVariant;
 }
@@ -101,7 +101,7 @@ function InteractiveContainer({
   border,
   roundingVariant = "default",
   heightVariant = "lg",
-  widthVariant = "auto",
+  widthVariant = "fit",
   ...props
 }: InteractiveContainerProps) {
   const { allowClick } = useDisabled();

web/lib/opal/src/core/interactive/stateful/styles.css

@@ -59,8 +59,8 @@
    --------------------------------------------------------------------------- */
 .interactive[data-interactive-variant="select-heavy"][data-interactive-state="filled"] {
   @apply bg-transparent;
-  --interactive-foreground: var(--text-04);
-  --interactive-foreground-icon: var(--text-04);
+  --interactive-foreground: var(--action-link-05);
+  --interactive-foreground-icon: var(--action-link-05);
 }
 .interactive[data-interactive-variant="select-heavy"][data-interactive-state="filled"]:hover:not(
     [data-disabled]
@@ -76,9 +76,7 @@
 .interactive[data-interactive-variant="select-heavy"][data-interactive-state="filled"][data-interaction="active"]:not(
     [data-disabled]
   ) {
-  @apply bg-background-neutral-00;
-  --interactive-foreground: var(--text-05);
-  --interactive-foreground-icon: var(--text-05);
+  @apply bg-background-tint-00;
 }
 .interactive[data-interactive-variant="select-heavy"][data-interactive-state="filled"][data-disabled] {
   @apply bg-transparent;
@@ -158,8 +156,8 @@
    --------------------------------------------------------------------------- */
 .interactive[data-interactive-variant="select-light"][data-interactive-state="filled"] {
   @apply bg-transparent;
-  --interactive-foreground: var(--text-04);
-  --interactive-foreground-icon: var(--text-04);
+  --interactive-foreground: var(--action-link-05);
+  --interactive-foreground-icon: var(--action-link-05);
 }
 .interactive[data-interactive-variant="select-light"][data-interactive-state="filled"]:hover:not(
     [data-disabled]
@@ -175,9 +173,7 @@
 .interactive[data-interactive-variant="select-light"][data-interactive-state="filled"][data-interaction="active"]:not(
     [data-disabled]
   ) {
-  @apply bg-background-neutral-00;
-  --interactive-foreground: var(--text-05);
-  --interactive-foreground-icon: var(--text-05);
+  @apply bg-background-tint-00;
 }
 .interactive[data-interactive-variant="select-light"][data-interactive-state="filled"][data-disabled] {
   @apply bg-transparent;

web/lib/opal/src/layouts/content/BodyLayout.tsx

@@ -40,6 +40,9 @@ interface BodyLayoutProps {
 
   /** Title prominence. Default: `"default"`. */
   prominence?: BodyProminence;
+
+  /** Ref forwarded to the root `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -80,6 +83,7 @@ function BodyLayout({
   sizePreset = "main-ui",
   orientation = "inline",
   prominence = "default",
+  ref,
 }: BodyLayoutProps) {
   const config = BODY_PRESETS[sizePreset];
   const titleColorClass =
@@ -87,6 +91,7 @@ function BodyLayout({
 
   return (
     <div
+      ref={ref}
       className="opal-content-body"
       data-orientation={orientation}
       style={{ gap: config.gap }}

web/lib/opal/src/layouts/content/ContentLg.tsx

@@ -48,6 +48,12 @@ interface ContentLgProps {
 
   /** Size preset. Default: `"headline"`. */
   sizePreset?: ContentLgSizePreset;
+
+  /** When `true`, the title color hooks into `Interactive`'s `--interactive-foreground` variable. */
+  withInteractive?: boolean;
+
+  /** Ref forwarded to the root `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -86,6 +92,8 @@ function ContentLg({
   description,
   editable,
   onTitleChange,
+  withInteractive,
+  ref,
 }: ContentLgProps) {
   const [editing, setEditing] = useState(false);
   const [editValue, setEditValue] = useState(title);
@@ -104,7 +112,12 @@ function ContentLg({
   }
 
   return (
-    <div className="opal-content-lg" style={{ gap: config.gap }}>
+    <div
+      ref={ref}
+      className="opal-content-lg"
+      data-interactive={withInteractive || undefined}
+      style={{ gap: config.gap }}
+    >
       {Icon && (
         <div
           className={cn(

web/lib/opal/src/layouts/content/ContentMd.tsx

@@ -61,6 +61,12 @@ interface ContentMdProps {
 
   /** Size preset. Default: `"main-ui"`. */
   sizePreset?: ContentMdSizePreset;
+
+  /** When `true`, the title color hooks into `Interactive`'s `--interactive-foreground` variable. */
+  withInteractive?: boolean;
+
+  /** Ref forwarded to the root `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -130,6 +136,8 @@ function ContentMd({
   auxIcon,
   tag,
   sizePreset = "main-ui",
+  withInteractive,
+  ref,
 }: ContentMdProps) {
   const [editing, setEditing] = useState(false);
   const [editValue, setEditValue] = useState(title);
@@ -149,7 +157,12 @@ function ContentMd({
   }
 
   return (
-    <div className="opal-content-md" style={{ gap: config.gap }}>
+    <div
+      ref={ref}
+      className="opal-content-md"
+      data-interactive={withInteractive || undefined}
+      style={{ gap: config.gap }}
+    >
       {Icon && (
         <div
           className={cn(

web/lib/opal/src/layouts/content/ContentSm.tsx

@@ -40,6 +40,12 @@ interface ContentSmProps {
 
   /** Title prominence. Default: `"default"`. */
   prominence?: ContentSmProminence;
+
+  /** When `true`, the title color hooks into `Interactive`'s `--interactive-foreground` variable. */
+  withInteractive?: boolean;
+
+  /** Ref forwarded to the root `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -80,14 +86,18 @@ function ContentSm({
   sizePreset = "main-ui",
   orientation = "inline",
   prominence = "default",
+  withInteractive,
+  ref,
 }: ContentSmProps) {
   const config = CONTENT_SM_PRESETS[sizePreset];
 
   return (
     <div
+      ref={ref}
       className="opal-content-sm"
       data-orientation={orientation}
       data-prominence={prominence}
+      data-interactive={withInteractive || undefined}
       style={{ gap: config.gap }}
     >
       {Icon && (

web/lib/opal/src/layouts/content/ContentXl.tsx

@@ -60,6 +60,12 @@ interface ContentXlProps {
 
   /** Optional tertiary icon rendered in the icon row. */
   moreIcon2?: IconFunctionComponent;
+
+  /** When `true`, the title color hooks into `Interactive`'s `--interactive-foreground` variable. */
+  withInteractive?: boolean;
+
+  /** Ref forwarded to the root `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -106,6 +112,8 @@ function ContentXl({
   onTitleChange,
   moreIcon1: MoreIcon1,
   moreIcon2: MoreIcon2,
+  withInteractive,
+  ref,
 }: ContentXlProps) {
   const [editing, setEditing] = useState(false);
   const [editValue, setEditValue] = useState(title);
@@ -124,7 +132,11 @@ function ContentXl({
   }
 
   return (
-    <div className="opal-content-xl">
+    <div
+      ref={ref}
+      className="opal-content-xl"
+      data-interactive={withInteractive || undefined}
+    >
       {(Icon || MoreIcon1 || MoreIcon2) && (
         <div className="opal-content-xl-icon-row">
           {Icon && (

web/lib/opal/src/layouts/content/HeadingLayout.tsx

@@ -52,6 +52,9 @@ interface HeadingLayoutProps {
 
   /** Variant controls icon placement. `"heading"` = top, `"section"` = inline. Default: `"heading"`. */
   variant?: HeadingVariant;
+
+  /** Ref forwarded to the root `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -91,6 +94,7 @@ function HeadingLayout({
   description,
   editable,
   onTitleChange,
+  ref,
 }: HeadingLayoutProps) {
   const [editing, setEditing] = useState(false);
   const [editValue, setEditValue] = useState(title);
@@ -112,6 +116,7 @@ function HeadingLayout({
 
   return (
     <div
+      ref={ref}
       className="opal-content-heading"
       data-icon-placement={iconPlacement}
       style={{ gap: iconPlacement === "left" ? config.gap : undefined }}

web/lib/opal/src/layouts/content/LabelLayout.tsx

@@ -61,6 +61,9 @@ interface LabelLayoutProps {
 
   /** Size preset. Default: `"main-ui"`. */
   sizePreset?: LabelSizePreset;
+
+  /** Ref forwarded to the root `<div>`. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -130,6 +133,7 @@ function LabelLayout({
   auxIcon,
   tag,
   sizePreset = "main-ui",
+  ref,
 }: LabelLayoutProps) {
   const [editing, setEditing] = useState(false);
   const [editValue, setEditValue] = useState(title);
@@ -149,7 +153,7 @@ function LabelLayout({
   }
 
   return (
-    <div className="opal-content-label" style={{ gap: config.gap }}>
+    <div ref={ref} className="opal-content-label" style={{ gap: config.gap }}>
       {Icon && (
         <div
           className={cn(

web/lib/opal/src/layouts/content/components.tsx

@@ -54,11 +54,18 @@ interface ContentBaseProps {
    * Uses the shared `WidthVariant` scale from `@opal/shared`.
    *
    * - `"auto"` — Shrink-wraps to content width
+   * - `"fit"` — Shrink-wraps to content width
    * - `"full"` — Stretches to fill the parent's width
    *
-   * @default "auto"
+   * @default "fit"
    */
   widthVariant?: WidthVariant;
+
+  /** When `true`, the title color hooks into `Interactive.Stateful`/`Interactive.Stateless`'s `--interactive-foreground` variable. */
+  withInteractive?: boolean;
+
+  /** Ref forwarded to the root `<div>` of the resolved layout. */
+  ref?: React.Ref<HTMLDivElement>;
 }
 
 // ---------------------------------------------------------------------------
@@ -122,11 +129,11 @@ function Content(props: ContentProps) {
     sizePreset = "headline",
     variant = "heading",
     widthVariant = "auto",
+    withInteractive,
+    ref,
     ...rest
   } = props;
 
-  const widthClass = widthVariants[widthVariant];
-
   let layout: React.ReactNode = null;
 
   // ContentXl / ContentLg: headline/section presets
@@ -135,6 +142,8 @@ function Content(props: ContentProps) {
       layout = (
         <ContentXl
           sizePreset={sizePreset}
+          withInteractive={withInteractive}
+          ref={ref}
           {...(rest as Omit<ContentXlProps, "sizePreset">)}
         />
       );
@@ -142,6 +151,8 @@ function Content(props: ContentProps) {
       layout = (
         <ContentLg
           sizePreset={sizePreset}
+          withInteractive={withInteractive}
+          ref={ref}
           {...(rest as Omit<ContentLgProps, "sizePreset">)}
         />
       );
@@ -154,6 +165,8 @@ function Content(props: ContentProps) {
     layout = (
       <ContentMd
         sizePreset={sizePreset}
+        withInteractive={withInteractive}
+        ref={ref}
         {...(rest as Omit<ContentMdProps, "sizePreset">)}
       />
     );
@@ -164,6 +177,8 @@ function Content(props: ContentProps) {
     layout = (
       <ContentSm
         sizePreset={sizePreset}
+        withInteractive={withInteractive}
+        ref={ref}
         {...(rest as Omit<
           React.ComponentProps<typeof ContentSm>,
           "sizePreset"
@@ -178,11 +193,7 @@ function Content(props: ContentProps) {
       `Content: no layout matched for sizePreset="${sizePreset}" variant="${variant}"`
     );
 
-  // "auto" → return layout directly (a block div with w-auto still
-  // stretches to its parent, defeating shrink-to-content).
-  if (widthVariant === "auto") return layout;
-
-  return <div className={widthClass}>{layout}</div>;
+  return <div className={widthVariants[widthVariant]}>{layout}</div>;
 }
 
 // ---------------------------------------------------------------------------

web/lib/opal/src/layouts/content/styles.css

@@ -22,6 +22,7 @@
 
 .opal-content-xl-icon-row {
   @apply flex flex-row items-center;
+  gap: 0.25rem;
 }
 
 /* ---------------------------------------------------------------------------
@@ -63,7 +64,7 @@
 }
 
 .opal-content-xl-title {
-  @apply text-left overflow-hidden;
+  @apply text-left overflow-hidden text-text-04;
   display: -webkit-box;
   -webkit-box-orient: vertical;
   -webkit-line-clamp: 1;
@@ -162,7 +163,7 @@
 }
 
 .opal-content-lg-title {
-  @apply text-left overflow-hidden;
+  @apply text-left overflow-hidden text-text-04;
   display: -webkit-box;
   -webkit-box-orient: vertical;
   -webkit-line-clamp: 1;
@@ -255,7 +256,7 @@
 }
 
 .opal-content-md-title {
-  @apply text-left overflow-hidden;
+  @apply text-left overflow-hidden text-text-04;
   display: -webkit-box;
   -webkit-box-orient: vertical;
   -webkit-line-clamp: 1;
@@ -361,8 +362,12 @@
   @apply text-text-03;
 }
 
+.opal-content-sm[data-prominence="muted"] .opal-content-sm-icon {
+  @apply text-text-02;
+}
+
 .opal-content-sm[data-prominence="muted-2x"] .opal-content-sm-icon {
-  @apply text-text-02 stroke-text-02;
+  @apply text-text-02;
 }
 
 /* ---------------------------------------------------------------------------
@@ -385,3 +390,44 @@
 .opal-content-sm[data-prominence="muted-2x"] .opal-content-sm-title {
   @apply text-text-02;
 }
+
+/* ===========================================================================
+   Interactive-foreground opt-in
+
+   When a Content variant is nested inside an Interactive and
+   `withInteractive` is set, the title and icon delegate their color to the
+   `--interactive-foreground` / `--interactive-foreground-icon` CSS variables
+   controlled by the ancestor Interactive variant.
+   =========================================================================== */
+
+.opal-content-xl[data-interactive] .opal-content-xl-title {
+  color: var(--interactive-foreground);
+}
+
+.opal-content-xl[data-interactive] .opal-content-xl-icon {
+  color: var(--interactive-foreground-icon);
+}
+
+.opal-content-lg[data-interactive] .opal-content-lg-title {
+  color: var(--interactive-foreground);
+}
+
+.opal-content-lg[data-interactive] .opal-content-lg-icon {
+  color: var(--interactive-foreground-icon);
+}
+
+.opal-content-md[data-interactive] .opal-content-md-title {
+  color: var(--interactive-foreground);
+}
+
+.opal-content-md[data-interactive] .opal-content-md-icon {
+  color: var(--interactive-foreground-icon);
+}
+
+.opal-content-sm[data-interactive] .opal-content-sm-title {
+  color: var(--interactive-foreground);
+}
+
+.opal-content-sm[data-interactive] .opal-content-sm-icon {
+  color: var(--interactive-foreground-icon);
+}

web/lib/opal/src/shared.ts

@@ -67,10 +67,12 @@ type SizeVariant = keyof typeof sizeVariants;
  * | Key    | Tailwind class |
  * |--------|----------------|
  * | `auto` | `w-auto`       |
+ * | `fit`  | `w-fit`        |
  * | `full` | `w-full`       |
  */
 const widthVariants = {
   auto: "w-auto",
+  fit: "w-fit",
   full: "w-full",
 } as const;
 

web/lib/opal/src/types.ts

@@ -30,6 +30,11 @@ export interface IconProps extends SVGProps<SVGSVGElement> {
 /** Strips `className` and `style` from a props type to enforce design-system styling. */
 export type WithoutStyles<T> = Omit<T, "className" | "style">;
 
+/** Like `Omit` but distributes over union types, preserving discriminated unions. */
+export type DistributiveOmit<T, K extends keyof any> = T extends any
+  ? Omit<T, K>
+  : never;
+
 /**
  * A React function component that accepts {@link IconProps}.
  *

web/next.config.js

@@ -8,7 +8,8 @@ const cspHeader = `
     base-uri 'self';
     form-action 'self';
     ${
-      process.env.NEXT_PUBLIC_CLOUD_ENABLED === "true"
+      process.env.NEXT_PUBLIC_CLOUD_ENABLED === "true" &&
+      process.env.NODE_ENV !== "development"
         ? "upgrade-insecure-requests;"
         : ""
     }

web/src/app/admin/bots/SlackBotCreationForm.tsx

@@ -7,7 +7,7 @@ import { SlackTokensForm } from "./SlackTokensForm";
 import * as SettingsLayouts from "@/layouts/settings-layouts";
 import { SvgSlack } from "@opal/icons";
 
-export const NewSlackBotForm = () => {
+export function NewSlackBotForm() {
   const [formValues] = useState({
     name: "",
     enabled: true,
@@ -19,7 +19,12 @@ export const NewSlackBotForm = () => {
 
   return (
     <SettingsLayouts.Root>
-      <SettingsLayouts.Header icon={SvgSlack} title="New Slack Bot" separator />
+      <SettingsLayouts.Header
+        icon={SvgSlack}
+        title="New Slack Bot"
+        separator
+        backButton
+      />
       <SettingsLayouts.Body>
         <CardSection>
           <div className="p-4">
@@ -33,4 +38,4 @@ export const NewSlackBotForm = () => {
       </SettingsLayouts.Body>
     </SettingsLayouts.Root>
   );
-};
+}

web/src/app/admin/bots/SlackBotUpdateForm.tsx

@@ -1,12 +1,12 @@
 "use client";
 
 import { toast } from "@/hooks/useToast";
-import { SlackBot, ValidSources } from "@/lib/types";
+import { SlackBot } from "@/lib/types";
 import { useRouter } from "next/navigation";
 import { useState, useEffect, useRef } from "react";
 import { updateSlackBotField } from "@/lib/updateSlackBotField";
 import { SlackTokensForm } from "./SlackTokensForm";
-import { SourceIcon } from "@/components/SourceIcon";
+
 import { EditableStringFieldDisplay } from "@/components/EditableStringFieldDisplay";
 import { deleteSlackBot } from "./new/lib";
 import GenericConfirmModal from "@/components/modals/GenericConfirmModal";
@@ -90,10 +90,7 @@ export const ExistingSlackBotForm = ({
     <div>
       <div className="flex items-center justify-between h-14">
         <div className="flex items-center gap-2">
-          <div className="my-auto">
-            <SourceIcon iconSize={32} sourceType={ValidSources.Slack} />
-          </div>
-          <div className="ml-1">
+          <div>
             <EditableStringFieldDisplay
               value={formValues.name}
               isEditable={true}

web/src/app/admin/bots/[bot-id]/channels/[id]/page.tsx

@@ -1,100 +1,122 @@
-import { SlackChannelConfigCreationForm } from "../SlackChannelConfigCreationForm";
-import { fetchSS } from "@/lib/utilsSS";
+"use client";
+
+import { use } from "react";
+import { SlackChannelConfigCreationForm } from "@/app/admin/bots/[bot-id]/channels/SlackChannelConfigCreationForm";
 import { ErrorCallout } from "@/components/ErrorCallout";
-import { DocumentSetSummary, SlackChannelConfig } from "@/lib/types";
-import { InstantSSRAutoRefresh } from "@/components/SSRAutoRefresh";
+import SimpleLoader from "@/refresh-components/loaders/SimpleLoader";
 import * as SettingsLayouts from "@/layouts/settings-layouts";
 import { SvgSlack } from "@opal/icons";
-import { FetchAgentsResponse, fetchAgentsSS } from "@/lib/agentsSS";
-import { getStandardAnswerCategoriesIfEE } from "@/components/standardAnswers/getStandardAnswerCategoriesIfEE";
+import { useSlackChannelConfigs } from "@/app/admin/bots/[bot-id]/hooks";
+import { useDocumentSets } from "@/app/admin/documents/sets/hooks";
+import { useAgents } from "@/hooks/useAgents";
+import { useStandardAnswerCategories } from "@/app/ee/admin/standard-answer/hooks";
+import { usePaidEnterpriseFeaturesEnabled } from "@/components/settings/usePaidEnterpriseFeaturesEnabled";
+import type { StandardAnswerCategoryResponse } from "@/components/standardAnswers/getStandardAnswerCategoriesIfEE";
 
-async function EditslackChannelConfigPage(props: {
-  params: Promise<{ id: number }>;
-}) {
-  const params = await props.params;
-  const tasks = [
-    fetchSS("/manage/admin/slack-app/channel"),
-    fetchSS("/manage/document-set"),
-    fetchAgentsSS(),
-  ];
+function EditSlackChannelConfigContent({ id }: { id: string }) {
+  const isPaidEnterprise = usePaidEnterpriseFeaturesEnabled();
 
-  const [
-    slackChannelsResponse,
-    documentSetsResponse,
-    [assistants, agentsFetchError],
-  ] = (await Promise.all(tasks)) as [Response, Response, FetchAgentsResponse];
+  const {
+    data: slackChannelConfigs,
+    isLoading: isChannelsLoading,
+    error: channelsError,
+  } = useSlackChannelConfigs();
 
-  const eeStandardAnswerCategoryResponse =
-    await getStandardAnswerCategoriesIfEE();
+  const {
+    data: documentSets,
+    isLoading: isDocSetsLoading,
+    error: docSetsError,
+  } = useDocumentSets();
 
-  if (!slackChannelsResponse.ok) {
-    return (
-      <ErrorCallout
-        errorTitle="Something went wrong :("
-        errorMsg={`Failed to fetch Slack Channels - ${await slackChannelsResponse.text()}`}
-      />
-    );
-  }
-  const allslackChannelConfigs =
-    (await slackChannelsResponse.json()) as SlackChannelConfig[];
+  const {
+    agents,
+    isLoading: isAgentsLoading,
+    error: agentsError,
+  } = useAgents();
 
-  const slackChannelConfig = allslackChannelConfigs.find(
-    (config) => config.id === Number(params.id)
+  const {
+    data: standardAnswerCategories,
+    isLoading: isStdAnswerLoading,
+    error: stdAnswerError,
+  } = useStandardAnswerCategories();
+
+  const isLoading =
+    isChannelsLoading ||
+    isDocSetsLoading ||
+    isAgentsLoading ||
+    (isPaidEnterprise && isStdAnswerLoading);
+
+  const slackChannelConfig = slackChannelConfigs?.find(
+    (config) => config.id === Number(id)
   );
 
-  if (!slackChannelConfig) {
-    return (
-      <ErrorCallout
-        errorTitle="Something went wrong :("
-        errorMsg={`Did not find Slack Channel config with ID: ${params.id}`}
-      />
-    );
-  }
-
-  if (!documentSetsResponse.ok) {
-    return (
-      <ErrorCallout
-        errorTitle="Something went wrong :("
-        errorMsg={`Failed to fetch document sets - ${await documentSetsResponse.text()}`}
-      />
-    );
-  }
-  const response = await documentSetsResponse.json();
-  const documentSets = response as DocumentSetSummary[];
-
-  if (agentsFetchError) {
-    return (
-      <ErrorCallout
-        errorTitle="Something went wrong :("
-        errorMsg={`Failed to fetch personas - ${agentsFetchError}`}
-      />
-    );
-  }
+  const title = slackChannelConfig?.is_default
+    ? "Edit Default Slack Config"
+    : "Edit Slack Channel Config";
 
   return (
     <SettingsLayouts.Root>
-      <InstantSSRAutoRefresh />
       <SettingsLayouts.Header
         icon={SvgSlack}
-        title={
-          slackChannelConfig.is_default
-            ? "Edit Default Slack Config"
-            : "Edit Slack Channel Config"
-        }
+        title={title}
         separator
         backButton
       />
       <SettingsLayouts.Body>
-        <SlackChannelConfigCreationForm
-          slack_bot_id={slackChannelConfig.slack_bot_id}
-          documentSets={documentSets}
-          personas={assistants}
-          standardAnswerCategoryResponse={eeStandardAnswerCategoryResponse}
-          existingSlackChannelConfig={slackChannelConfig}
-        />
+        {isLoading ? (
+          <SimpleLoader />
+        ) : channelsError || !slackChannelConfigs ? (
+          <ErrorCallout
+            errorTitle="Something went wrong :("
+            errorMsg={`Failed to fetch Slack Channels - ${
+              channelsError?.message ?? "unknown error"
+            }`}
+          />
+        ) : !slackChannelConfig ? (
+          <ErrorCallout
+            errorTitle="Something went wrong :("
+            errorMsg={`Did not find Slack Channel config with ID: ${id}`}
+          />
+        ) : docSetsError || !documentSets ? (
+          <ErrorCallout
+            errorTitle="Something went wrong :("
+            errorMsg={`Failed to fetch document sets - ${
+              docSetsError?.message ?? "unknown error"
+            }`}
+          />
+        ) : agentsError ? (
+          <ErrorCallout
+            errorTitle="Something went wrong :("
+            errorMsg={`Failed to fetch agents - ${
+              agentsError?.message ?? "unknown error"
+            }`}
+          />
+        ) : (
+          <SlackChannelConfigCreationForm
+            slack_bot_id={slackChannelConfig.slack_bot_id}
+            documentSets={documentSets}
+            personas={agents}
+            standardAnswerCategoryResponse={
+              isPaidEnterprise
+                ? {
+                    paidEnterpriseFeaturesEnabled: true,
+                    categories: standardAnswerCategories ?? [],
+                    ...(stdAnswerError
+                      ? { error: { message: String(stdAnswerError) } }
+                      : {}),
+                  }
+                : { paidEnterpriseFeaturesEnabled: false }
+            }
+            existingSlackChannelConfig={slackChannelConfig}
+          />
+        )}
       </SettingsLayouts.Body>
     </SettingsLayouts.Root>
   );
 }
 
-export default EditslackChannelConfigPage;
+export default function Page(props: { params: Promise<{ id: string }> }) {
+  const params = use(props.params);
+
+  return <EditSlackChannelConfigContent id={params.id} />;
+}

web/src/app/admin/bots/[bot-id]/channels/new/page.tsx

@@ -1,53 +1,109 @@
-import { SlackChannelConfigCreationForm } from "../SlackChannelConfigCreationForm";
-import { fetchSS } from "@/lib/utilsSS";
+"use client";
+
+import { use, useEffect } from "react";
+import { SlackChannelConfigCreationForm } from "@/app/admin/bots/[bot-id]/channels/SlackChannelConfigCreationForm";
 import { ErrorCallout } from "@/components/ErrorCallout";
-import { DocumentSetSummary } from "@/lib/types";
-import { fetchAgentsSS } from "@/lib/agentsSS";
-import { getStandardAnswerCategoriesIfEE } from "@/components/standardAnswers/getStandardAnswerCategoriesIfEE";
-import { redirect } from "next/navigation";
+import SimpleLoader from "@/refresh-components/loaders/SimpleLoader";
 import * as SettingsLayouts from "@/layouts/settings-layouts";
 import { SvgSlack } from "@opal/icons";
+import { useDocumentSets } from "@/app/admin/documents/sets/hooks";
+import { useAgents } from "@/hooks/useAgents";
+import { useStandardAnswerCategories } from "@/app/ee/admin/standard-answer/hooks";
+import { usePaidEnterpriseFeaturesEnabled } from "@/components/settings/usePaidEnterpriseFeaturesEnabled";
+import type { StandardAnswerCategoryResponse } from "@/components/standardAnswers/getStandardAnswerCategoriesIfEE";
+import { useRouter } from "next/navigation";
+
+function NewChannelConfigContent({ slackBotId }: { slackBotId: number }) {
+  const isPaidEnterprise = usePaidEnterpriseFeaturesEnabled();
+
+  const {
+    data: documentSets,
+    isLoading: isDocSetsLoading,
+    error: docSetsError,
+  } = useDocumentSets();
+
+  const {
+    agents,
+    isLoading: isAgentsLoading,
+    error: agentsError,
+  } = useAgents();
+
+  const {
+    data: standardAnswerCategories,
+    isLoading: isStdAnswerLoading,
+    error: stdAnswerError,
+  } = useStandardAnswerCategories();
+
+  if (
+    isDocSetsLoading ||
+    isAgentsLoading ||
+    (isPaidEnterprise && isStdAnswerLoading)
+  ) {
+    return <SimpleLoader />;
+  }
+
+  if (docSetsError || !documentSets) {
+    return (
+      <ErrorCallout
+        errorTitle="Something went wrong :("
+        errorMsg={`Failed to fetch document sets - ${
+          docSetsError?.message ?? "unknown error"
+        }`}
+      />
+    );
+  }
+
+  if (agentsError) {
+    return (
+      <ErrorCallout
+        errorTitle="Something went wrong :("
+        errorMsg={`Failed to fetch agents - ${
+          agentsError?.message ?? "unknown error"
+        }`}
+      />
+    );
+  }
+
+  const standardAnswerCategoryResponse: StandardAnswerCategoryResponse =
+    isPaidEnterprise
+      ? {
+          paidEnterpriseFeaturesEnabled: true,
+          categories: standardAnswerCategories ?? [],
+          ...(stdAnswerError
+            ? { error: { message: String(stdAnswerError) } }
+            : {}),
+        }
+      : { paidEnterpriseFeaturesEnabled: false };
+
+  return (
+    <SlackChannelConfigCreationForm
+      slack_bot_id={slackBotId}
+      documentSets={documentSets}
+      personas={agents}
+      standardAnswerCategoryResponse={standardAnswerCategoryResponse}
+    />
+  );
+}
+
+export default function Page(props: { params: Promise<{ "bot-id": string }> }) {
+  const unwrappedParams = use(props.params);
+  const router = useRouter();
 
-async function NewChannelConfigPage(props: {
-  params: Promise<{ "bot-id": string }>;
-}) {
-  const unwrappedParams = await props.params;
   const slack_bot_id_raw = unwrappedParams?.["bot-id"] || null;
   const slack_bot_id = slack_bot_id_raw
     ? parseInt(slack_bot_id_raw as string, 10)
     : null;
+
+  useEffect(() => {
+    if (!slack_bot_id || isNaN(slack_bot_id)) {
+      router.replace("/admin/bots");
+    }
+  }, [slack_bot_id, router]);
+
   if (!slack_bot_id || isNaN(slack_bot_id)) {
-    redirect("/admin/bots");
     return null;
   }
 
-  const [documentSetsResponse, agentsResponse, standardAnswerCategoryResponse] =
-    await Promise.all([
-      fetchSS("/manage/document-set") as Promise<Response>,
-      fetchAgentsSS(),
-      getStandardAnswerCategoriesIfEE(),
-    ]);
-
-  if (!documentSetsResponse.ok) {
-    return (
-      <ErrorCallout
-        errorTitle="Something went wrong :("
-        errorMsg={`Failed to fetch document sets - ${await documentSetsResponse.text()}`}
-      />
-    );
-  }
-  const documentSets =
-    (await documentSetsResponse.json()) as DocumentSetSummary[];
-
-  if (agentsResponse[1]) {
-    return (
-      <ErrorCallout
-        errorTitle="Something went wrong :("
-        errorMsg={`Failed to fetch agents - ${agentsResponse[1]}`}
-      />
-    );
-  }
-
   return (
     <SettingsLayouts.Root>
       <SettingsLayouts.Header
@@ -57,15 +113,8 @@ async function NewChannelConfigPage(props: {
         backButton
       />
       <SettingsLayouts.Body>
-        <SlackChannelConfigCreationForm
-          slack_bot_id={slack_bot_id}
-          documentSets={documentSets}
-          personas={agentsResponse[0]}
-          standardAnswerCategoryResponse={standardAnswerCategoryResponse}
-        />
+        <NewChannelConfigContent slackBotId={slack_bot_id} />
       </SettingsLayouts.Body>
     </SettingsLayouts.Root>
   );
 }
-
-export default NewChannelConfigPage;

web/src/app/admin/bots/[bot-id]/page.tsx

@@ -1,82 +1,62 @@
 "use client";
 
 import { use } from "react";
-import BackButton from "@/refresh-components/buttons/BackButton";
 import { ErrorCallout } from "@/components/ErrorCallout";
-import { ThreeDotsLoader } from "@/components/Loading";
-import { InstantSSRAutoRefresh } from "@/components/SSRAutoRefresh";
+import SimpleLoader from "@/refresh-components/loaders/SimpleLoader";
 import SlackChannelConfigsTable from "./SlackChannelConfigsTable";
 import { useSlackBot, useSlackChannelConfigsByBot } from "./hooks";
 import { ExistingSlackBotForm } from "../SlackBotUpdateForm";
-import Separator from "@/refresh-components/Separator";
-
-function SlackBotEditPage({
-  params,
-}: {
-  params: Promise<{ "bot-id": string }>;
-}) {
-  // Unwrap the params promise
-  const unwrappedParams = use(params);
+import * as SettingsLayouts from "@/layouts/settings-layouts";
+import { SvgSlack } from "@opal/icons";
+import { getErrorMsg } from "@/lib/error";
 
+function SlackBotEditContent({ botId }: { botId: string }) {
   const {
     data: slackBot,
     isLoading: isSlackBotLoading,
     error: slackBotError,
     refreshSlackBot,
-  } = useSlackBot(Number(unwrappedParams["bot-id"]));
+  } = useSlackBot(Number(botId));
 
   const {
     data: slackChannelConfigs,
     isLoading: isSlackChannelConfigsLoading,
     error: slackChannelConfigsError,
     refreshSlackChannelConfigs,
-  } = useSlackChannelConfigsByBot(Number(unwrappedParams["bot-id"]));
+  } = useSlackChannelConfigsByBot(Number(botId));
 
   if (isSlackBotLoading || isSlackChannelConfigsLoading) {
-    return (
-      <div className="flex justify-center items-center h-screen">
-        <ThreeDotsLoader />
-      </div>
-    );
+    return <SimpleLoader />;
   }
 
   if (slackBotError || !slackBot) {
-    const errorMsg =
-      slackBotError?.info?.message ||
-      slackBotError?.info?.detail ||
-      "An unknown error occurred";
     return (
       <ErrorCallout
         errorTitle="Something went wrong :("
-        errorMsg={`Failed to fetch Slack Bot ${unwrappedParams["bot-id"]}: ${errorMsg}`}
+        errorMsg={`Failed to fetch Slack Bot ${botId}: ${getErrorMsg(
+          slackBotError
+        )}`}
       />
     );
   }
 
   if (slackChannelConfigsError || !slackChannelConfigs) {
-    const errorMsg =
-      slackChannelConfigsError?.info?.message ||
-      slackChannelConfigsError?.info?.detail ||
-      "An unknown error occurred";
     return (
       <ErrorCallout
         errorTitle="Something went wrong :("
-        errorMsg={`Failed to fetch Slack Bot ${unwrappedParams["bot-id"]}: ${errorMsg}`}
+        errorMsg={`Failed to fetch Slack Bot ${botId}: ${getErrorMsg(
+          slackChannelConfigsError
+        )}`}
       />
     );
   }
 
   return (
     <>
-      <InstantSSRAutoRefresh />
-
-      <BackButton routerOverride="/admin/bots" />
-
       <ExistingSlackBotForm
         existingSlackBot={slackBot}
         refreshSlackBot={refreshSlackBot}
       />
-      <Separator />
 
       <div className="mt-8">
         <SlackChannelConfigsTable
@@ -94,9 +74,19 @@ export default function Page({
 }: {
   params: Promise<{ "bot-id": string }>;
 }) {
+  const unwrappedParams = use(params);
+
   return (
-    <>
-      <SlackBotEditPage params={params} />
-    </>
+    <SettingsLayouts.Root>
+      <SettingsLayouts.Header
+        icon={SvgSlack}
+        title="Edit Slack Bot"
+        backButton
+        separator
+      />
+      <SettingsLayouts.Body>
+        <SlackBotEditContent botId={unwrappedParams["bot-id"]} />
+      </SettingsLayouts.Body>
+    </SettingsLayouts.Root>
   );
 }

web/src/app/admin/bots/new/page.tsx

@@ -1,12 +1,7 @@
-import BackButton from "@/refresh-components/buttons/BackButton";
+"use client";
+
 import { NewSlackBotForm } from "../SlackBotCreationForm";
 
-export default async function NewSlackBotPage() {
-  return (
-    <>
-      <BackButton routerOverride="/admin/bots" />
-
-      <NewSlackBotForm />
-    </>
-  );
+export default function Page() {
+  return <NewSlackBotForm />;
 }

web/src/app/app/message/messageComponents/renderMessageComponent.tsx

@@ -1,11 +1,14 @@
 import React, { JSX, memo } from "react";
 import {
   ChatPacket,
+  CODE_INTERPRETER_TOOL_TYPES,
   ImageGenerationToolPacket,
   Packet,
   PacketType,
   ReasoningPacket,
+  SearchToolStart,
   StopReason,
+  ToolCallArgumentDelta,
 } from "../../services/streamingModels";
 import {
   FullChatState,
@@ -26,7 +29,6 @@ import { DeepResearchPlanRenderer } from "./timeline/renderers/deepresearch/Deep
 import { ResearchAgentRenderer } from "./timeline/renderers/deepresearch/ResearchAgentRenderer";
 import { WebSearchToolRenderer } from "./timeline/renderers/search/WebSearchToolRenderer";
 import { InternalSearchToolRenderer } from "./timeline/renderers/search/InternalSearchToolRenderer";
-import { SearchToolStart } from "../../services/streamingModels";
 
 // Different types of chat packets using discriminated unions
 interface GroupedPackets {
@@ -56,7 +58,12 @@ function isImageToolPacket(packet: Packet) {
 }
 
 function isPythonToolPacket(packet: Packet) {
-  return packet.obj.type === PacketType.PYTHON_TOOL_START;
+  return (
+    packet.obj.type === PacketType.PYTHON_TOOL_START ||
+    (packet.obj.type === PacketType.TOOL_CALL_ARGUMENT_DELTA &&
+      (packet.obj as ToolCallArgumentDelta).tool_type ===
+        CODE_INTERPRETER_TOOL_TYPES.PYTHON)
+  );
 }
 
 function isCustomToolPacket(packet: Packet) {

web/src/app/app/message/messageComponents/timeline/hooks/packetProcessor.ts

@@ -10,6 +10,8 @@ import {
   Stop,
   ImageGenerationToolDelta,
   MessageStart,
+  ToolCallArgumentDelta,
+  CODE_INTERPRETER_TOOL_TYPES,
 } from "@/app/app/services/streamingModels";
 import { CitationMap } from "@/app/app/interfaces";
 import { OnyxDocument } from "@/lib/search/interfaces";
@@ -138,6 +140,7 @@ const CONTENT_PACKET_TYPES_SET = new Set<PacketType>([
   PacketType.SEARCH_TOOL_START,
   PacketType.IMAGE_GENERATION_TOOL_START,
   PacketType.PYTHON_TOOL_START,
+  PacketType.TOOL_CALL_ARGUMENT_DELTA,
   PacketType.CUSTOM_TOOL_START,
   PacketType.FILE_READER_START,
   PacketType.FETCH_TOOL_START,
@@ -149,9 +152,16 @@ const CONTENT_PACKET_TYPES_SET = new Set<PacketType>([
 ]);
 
 function hasContentPackets(packets: Packet[]): boolean {
-  return packets.some((packet) =>
-    CONTENT_PACKET_TYPES_SET.has(packet.obj.type as PacketType)
-  );
+  return packets.some((packet) => {
+    const type = packet.obj.type as PacketType;
+    if (type === PacketType.TOOL_CALL_ARGUMENT_DELTA) {
+      return (
+        (packet.obj as ToolCallArgumentDelta).tool_type ===
+        CODE_INTERPRETER_TOOL_TYPES.PYTHON
+      );
+    }
+    return CONTENT_PACKET_TYPES_SET.has(type);
+  });
 }
 
 /**

web/src/app/app/message/messageComponents/timeline/packetHelpers.ts

@@ -1,6 +1,13 @@
-import { Packet, PacketType } from "@/app/app/services/streamingModels";
+import {
+  CODE_INTERPRETER_TOOL_TYPES,
+  Packet,
+  PacketType,
+  ToolCallArgumentDelta,
+} from "@/app/app/services/streamingModels";
 
-// Packet types with renderers supporting collapsed streaming mode
+// Packet types with renderers supporting collapsed streaming mode.
+// TOOL_CALL_ARGUMENT_DELTA is intentionally excluded here because it requires
+// a tool_type check — it's handled separately in stepSupportsCollapsedStreaming.
 export const COLLAPSED_STREAMING_PACKET_TYPES = new Set<PacketType>([
   PacketType.SEARCH_TOOL_START,
   PacketType.FETCH_TOOL_START,
@@ -21,7 +28,13 @@ export const isSearchToolPackets = (packets: Packet[]): boolean =>
 
 // Check if packets belong to a python tool
 export const isPythonToolPackets = (packets: Packet[]): boolean =>
-  packets.some((p) => p.obj.type === PacketType.PYTHON_TOOL_START);
+  packets.some(
+    (p) =>
+      p.obj.type === PacketType.PYTHON_TOOL_START ||
+      (p.obj.type === PacketType.TOOL_CALL_ARGUMENT_DELTA &&
+        (p.obj as ToolCallArgumentDelta).tool_type ===
+          CODE_INTERPRETER_TOOL_TYPES.PYTHON)
+  );
 
 // Check if packets belong to reasoning
 export const isReasoningPackets = (packets: Packet[]): boolean =>
@@ -29,8 +42,12 @@ export const isReasoningPackets = (packets: Packet[]): boolean =>
 
 // Check if step supports collapsed streaming rendering mode
 export const stepSupportsCollapsedStreaming = (packets: Packet[]): boolean =>
-  packets.some((p) =>
-    COLLAPSED_STREAMING_PACKET_TYPES.has(p.obj.type as PacketType)
+  packets.some(
+    (p) =>
+      COLLAPSED_STREAMING_PACKET_TYPES.has(p.obj.type as PacketType) ||
+      (p.obj.type === PacketType.TOOL_CALL_ARGUMENT_DELTA &&
+        (p.obj as ToolCallArgumentDelta).tool_type ===
+          CODE_INTERPRETER_TOOL_TYPES.PYTHON)
   );
 
 // Check if packets have content worth rendering in collapsed streaming mode.
@@ -67,7 +84,13 @@ export const stepHasCollapsedStreamingContent = (
   // Python tool renders code/output from the start packet onward
   if (
     packetTypes.has(PacketType.PYTHON_TOOL_START) ||
-    packetTypes.has(PacketType.PYTHON_TOOL_DELTA)
+    packetTypes.has(PacketType.PYTHON_TOOL_DELTA) ||
+    packets.some(
+      (p) =>
+        p.obj.type === PacketType.TOOL_CALL_ARGUMENT_DELTA &&
+        (p.obj as ToolCallArgumentDelta).tool_type ===
+          CODE_INTERPRETER_TOOL_TYPES.PYTHON
+    )
   ) {
     return true;
   }

web/src/app/app/message/messageComponents/timeline/renderers/code/PythonToolRenderer.tsx

@@ -4,7 +4,9 @@ import {
   PythonToolPacket,
   PythonToolStart,
   PythonToolDelta,
+  ToolCallArgumentDelta,
   SectionEnd,
+  CODE_INTERPRETER_TOOL_TYPES,
 } from "@/app/app/services/streamingModels";
 import {
   MessageRenderer,
@@ -39,6 +41,18 @@ function HighlightedPythonCode({ code }: { code: string }) {
 
 // Helper function to construct current Python execution state
 function constructCurrentPythonState(packets: PythonToolPacket[]) {
+  // Accumulate streaming code from argument deltas (arrives before PythonToolStart)
+  const streamingCode = packets
+    .filter(
+      (packet) =>
+        packet.obj.type === PacketType.TOOL_CALL_ARGUMENT_DELTA &&
+        (packet.obj as ToolCallArgumentDelta).tool_type ===
+          CODE_INTERPRETER_TOOL_TYPES.PYTHON
+    )
+    .map((packet) =>
+      String((packet.obj as ToolCallArgumentDelta).argument_deltas.code ?? "")
+    )
+    .join("");
   const pythonStart = packets.find(
     (packet) => packet.obj.type === PacketType.PYTHON_TOOL_START
   )?.obj as PythonToolStart | null;
@@ -51,7 +65,8 @@ function constructCurrentPythonState(packets: PythonToolPacket[]) {
       packet.obj.type === PacketType.ERROR
   )?.obj as SectionEnd | null;
 
-  const code = pythonStart?.code || "";
+  // Use complete code from PythonToolStart if available, else use streamed code.
+  const code = pythonStart?.code || streamingCode;
   const stdout = pythonDeltas
     .map((delta) => delta?.stdout || "")
     .filter((s) => s)
@@ -61,6 +76,7 @@ function constructCurrentPythonState(packets: PythonToolPacket[]) {
     .filter((s) => s)
     .join("");
   const fileIds = pythonDeltas.flatMap((delta) => delta?.file_ids || []);
+  const isStreaming = !pythonStart && streamingCode.length > 0;
   const isExecuting = pythonStart && !pythonEnd;
   const isComplete = pythonStart && pythonEnd;
   const hasError = stderr.length > 0;
@@ -70,6 +86,7 @@ function constructCurrentPythonState(packets: PythonToolPacket[]) {
     stdout,
     stderr,
     fileIds,
+    isStreaming,
     isExecuting,
     isComplete,
     hasError,
@@ -82,8 +99,16 @@ export const PythonToolRenderer: MessageRenderer<PythonToolPacket, {}> = ({
   renderType,
   children,
 }) => {
-  const { code, stdout, stderr, fileIds, isExecuting, isComplete, hasError } =
-    constructCurrentPythonState(packets);
+  const {
+    code,
+    stdout,
+    stderr,
+    fileIds,
+    isStreaming,
+    isExecuting,
+    isComplete,
+    hasError,
+  } = constructCurrentPythonState(packets);
 
   useEffect(() => {
     if (isComplete) {
@@ -92,6 +117,9 @@ export const PythonToolRenderer: MessageRenderer<PythonToolPacket, {}> = ({
   }, [isComplete, onComplete]);
 
   const status = useMemo(() => {
+    if (isStreaming) {
+      return "Writing code...";
+    }
     if (isExecuting) {
       return "Executing Python code...";
     }
@@ -102,13 +130,13 @@ export const PythonToolRenderer: MessageRenderer<PythonToolPacket, {}> = ({
       return "Python execution completed";
     }
     return "Python execution";
-  }, [isComplete, isExecuting, hasError]);
+  }, [isStreaming, isComplete, isExecuting, hasError]);
 
   // Shared content for all states - used by both FULL and compact modes
   const content = (
     <div className="flex flex-col mb-1 space-y-2">
-      {/* Loading indicator when executing */}
-      {isExecuting && (
+      {/* Loading indicator when streaming or executing */}
+      {(isStreaming || isExecuting) && (
         <div className="flex items-center gap-2 text-sm text-muted-foreground">
           <div className="flex gap-0.5">
             <div className="w-1 h-1 bg-current rounded-full animate-pulse"></div>
@@ -121,7 +149,7 @@ export const PythonToolRenderer: MessageRenderer<PythonToolPacket, {}> = ({
               style={{ animationDelay: "0.2s" }}
             ></div>
           </div>
-          <span>Running code...</span>
+          <span>{isStreaming ? "Writing code..." : "Running code..."}</span>
         </div>
       )}
 

web/src/app/app/projects/projectsService.ts

@@ -308,15 +308,15 @@ export async function getProjectTokenCount(projectId: number): Promise<number> {
 
 export async function getMaxSelectedDocumentTokens(
   personaId: number
-): Promise<number> {
+): Promise<number | null> {
   const response = await fetch(
     `/api/chat/max-selected-document-tokens?persona_id=${personaId}`
   );
   if (!response.ok) {
-    return 128_000;
+    return null;
   }
   const json = await response.json();
-  return (json?.max_tokens as number) ?? 128_000;
+  return (json?.max_tokens as number) ?? null;
 }
 
 export async function moveChatSession(

web/src/app/app/services/lib.tsx

@@ -288,15 +288,15 @@ export async function deleteAllChatSessions() {
 
 export async function getAvailableContextTokens(
   chatSessionId: string
-): Promise<number> {
+): Promise<number | null> {
   const response = await fetch(
     `/api/chat/available-context-tokens/${chatSessionId}`
   );
   if (!response.ok) {
-    return 0;
+    return null;
   }
   const data = (await response.json()) as { available_tokens: number };
-  return data?.available_tokens ?? 0;
+  return data?.available_tokens ?? null;
 }
 
 export function processRawChatHistory(

web/src/app/app/services/packetUtils.ts

@@ -16,6 +16,7 @@ export function isToolPacket(
     PacketType.SEARCH_TOOL_DOCUMENTS_DELTA,
     PacketType.PYTHON_TOOL_START,
     PacketType.PYTHON_TOOL_DELTA,
+    PacketType.TOOL_CALL_ARGUMENT_DELTA,
     PacketType.CUSTOM_TOOL_START,
     PacketType.CUSTOM_TOOL_DELTA,
     PacketType.FILE_READER_START,

web/src/app/app/services/streamingModels.ts

@@ -27,6 +27,9 @@ export enum PacketType {
   FETCH_TOOL_URLS = "open_url_urls",
   FETCH_TOOL_DOCUMENTS = "open_url_documents",
 
+  // Tool call argument delta (streams tool args before tool executes)
+  TOOL_CALL_ARGUMENT_DELTA = "tool_call_argument_delta",
+
   // Custom tool packets
   CUSTOM_TOOL_START = "custom_tool_start",
   CUSTOM_TOOL_DELTA = "custom_tool_delta",
@@ -59,6 +62,10 @@ export enum PacketType {
   INTERMEDIATE_REPORT_CITED_DOCS = "intermediate_report_cited_docs",
 }
 
+export const CODE_INTERPRETER_TOOL_TYPES = {
+  PYTHON: "python",
+} as const;
+
 // Basic Message Packets
 export interface MessageStart extends BaseObj {
   id: string;
@@ -149,6 +156,13 @@ export interface PythonToolDelta extends BaseObj {
   file_ids: string[];
 }
 
+export interface ToolCallArgumentDelta extends BaseObj {
+  type: "tool_call_argument_delta";
+  tool_type: string;
+  tool_id: string;
+  argument_deltas: Record<string, unknown>;
+}
+
 export interface FetchToolStart extends BaseObj {
   type: "open_url_start";
 }
@@ -294,6 +308,7 @@ export type ImageGenerationToolObj =
 export type PythonToolObj =
   | PythonToolStart
   | PythonToolDelta
+  | ToolCallArgumentDelta
   | SectionEnd
   | PacketError;
 export type FetchToolObj =

web/src/hooks/useChatController.ts

@@ -2,9 +2,12 @@
 
 import {
   buildChatUrl,
+  getAvailableContextTokens,
   nameChatSession,
   updateLlmOverrideForChatSession,
 } from "@/app/app/services/lib";
+import { getMaxSelectedDocumentTokens } from "@/app/app/projects/projectsService";
+import { DEFAULT_CONTEXT_TOKENS } from "@/lib/constants";
 import { StreamStopInfo } from "@/lib/search/interfaces";
 import { useCallback, useEffect, useMemo, useRef, useState } from "react";
 import type { Route } from "next";
@@ -194,9 +197,6 @@ export default function useChatController({
 
   const navigatingAway = useRef(false);
 
-  // Local state that doesn't need to be in the store
-  const [_maxTokens, setMaxTokens] = useState<number>(4096);
-
   // Sync store state changes
   useEffect(() => {
     if (currentSessionId) {
@@ -1067,21 +1067,59 @@ export default function useChatController({
     handleSlackChatRedirect();
   }, [searchParams, router]);
 
-  // fetch # of allowed document tokens for the selected Persona
-  useEffect(() => {
-    if (!liveAgent?.id) return; // avoid calling with undefined persona id
+  // Available context tokens: if a chat session exists, fetch from the session
+  // API (dynamic per session/model). Otherwise derive from the persona's max
+  // document tokens. The backend already accounts for system prompt, tools,
+  // and user-message reservations.
+  const [availableContextTokens, setAvailableContextTokens] = useState<number>(
+    DEFAULT_CONTEXT_TOKENS
+  );
 
-    async function fetchMaxTokens() {
-      const response = await fetch(
-        `/api/chat/max-selected-document-tokens?persona_id=${liveAgent?.id}`
-      );
-      if (response.ok) {
-        const maxTokens = (await response.json()).max_tokens as number;
-        setMaxTokens(maxTokens);
+  useEffect(() => {
+    if (!llmManager.hasAnyProvider) return;
+
+    let cancelled = false;
+
+    const setIfActive = (tokens: number) => {
+      if (!cancelled) setAvailableContextTokens(tokens);
+    };
+
+    // Prefer the Zustand session ID, but fall back to the URL-derived prop
+    // so we don't incorrectly take the persona path while the store is
+    // still initialising on navigation to an existing chat.
+    const sessionId = currentSessionId || existingChatSessionId;
+
+    (async () => {
+      try {
+        if (sessionId) {
+          const available = await getAvailableContextTokens(sessionId);
+          setIfActive(available ?? DEFAULT_CONTEXT_TOKENS);
+          return;
+        }
+
+        const personaId = liveAgent?.id;
+        if (personaId == null) {
+          setIfActive(DEFAULT_CONTEXT_TOKENS);
+          return;
+        }
+
+        const maxTokens = await getMaxSelectedDocumentTokens(personaId);
+        setIfActive(maxTokens ?? DEFAULT_CONTEXT_TOKENS);
+      } catch (e) {
+        console.error("Failed to fetch available context tokens:", e);
+        setIfActive(DEFAULT_CONTEXT_TOKENS);
       }
-    }
-    fetchMaxTokens();
-  }, [liveAgent]);
+    })();
+
+    return () => {
+      cancelled = true;
+    };
+  }, [
+    currentSessionId,
+    existingChatSessionId,
+    liveAgent?.id,
+    llmManager.hasAnyProvider,
+  ]);
 
   // check if there's an image file in the message history so that we know
   // which LLMs are available to use
@@ -1110,5 +1148,7 @@ export default function useChatController({
     onSubmit,
     stopGenerating,
     handleMessageSpecificFileUpload,
+    // data
+    availableContextTokens,
   };
 }

web/src/interfaces/settings.ts

@@ -36,6 +36,7 @@ export interface Settings {
 
   // User Knowledge settings
   user_knowledge_enabled?: boolean;
+  user_file_max_upload_size_mb?: number | null;
 
   // Connector settings
   show_extra_connectors?: boolean;

web/src/lib/error.ts

@@ -0,0 +1,10 @@
+/**
+ * Extract a human-readable error message from an SWR error object.
+ * SWR errors from `errorHandlingFetcher` attach `info.message` or `info.detail`.
+ */
+export function getErrorMsg(
+  error: { info?: { message?: string; detail?: string } } | null | undefined,
+  fallback = "An unknown error occurred"
+): string {
+  return error?.info?.message || error?.info?.detail || fallback;
+}

web/src/providers/ProjectsContext.tsx

@@ -43,6 +43,7 @@ import { useAppRouter } from "@/hooks/appNavigation";
 import { ChatFileType } from "@/app/app/interfaces";
 import { toast } from "@/hooks/useToast";
 import { useProjects } from "@/lib/hooks/useProjects";
+import { SettingsContext } from "@/providers/SettingsProvider";
 
 export type { Project, ProjectFile } from "@/app/app/projects/projectsService";
 
@@ -84,6 +85,8 @@ function buildFileKey(file: File): string {
   return `${file.size}|${namePrefix}`;
 }
 
+const DEFAULT_USER_FILE_MAX_UPLOAD_SIZE_MB = 50;
+
 interface ProjectsContextType {
   projects: Project[];
   recentFiles: ProjectFile[];
@@ -157,6 +160,7 @@ export function ProjectsProvider({ children }: ProjectsProviderProps) {
     new Map()
   );
   const route = useAppRouter();
+  const settingsContext = useContext(SettingsContext);
 
   // Use SWR's mutate to refresh projects - returns the new data
   const fetchProjects = useCallback(async (): Promise<Project[]> => {
@@ -336,16 +340,40 @@ export function ProjectsProvider({ children }: ProjectsProviderProps) {
       onSuccess?: (uploaded: CategorizedFiles) => void,
       onFailure?: (failedTempIds: string[]) => void
     ): Promise<ProjectFile[]> => {
-      const optimisticFiles = files.map((f) =>
+      const rawMax = settingsContext?.settings?.user_file_max_upload_size_mb;
+      const maxUploadSizeMb =
+        rawMax && rawMax > 0 ? rawMax : DEFAULT_USER_FILE_MAX_UPLOAD_SIZE_MB;
+      const maxUploadSizeBytes = maxUploadSizeMb * 1024 * 1024;
+
+      const oversizedFiles = files.filter(
+        (file) => file.size > maxUploadSizeBytes
+      );
+      const validFiles = files.filter(
+        (file) => file.size <= maxUploadSizeBytes
+      );
+
+      if (oversizedFiles.length > 0) {
+        const skippedNames = oversizedFiles.map((file) => file.name).join(", ");
+        toast.warning(
+          `Skipped ${oversizedFiles.length} oversized file(s) (>${maxUploadSizeMb} MB): ${skippedNames}`
+        );
+      }
+
+      if (validFiles.length === 0) {
+        onFailure?.([]);
+        return [];
+      }
+
+      const optimisticFiles = validFiles.map((f) =>
         createOptimisticFile(f, projectId)
       );
-      const tempIdMap = getTempIdMap(files, optimisticFiles);
+      const tempIdMap = getTempIdMap(validFiles, optimisticFiles);
       setAllRecentFiles((prev) => [...optimisticFiles, ...prev]);
       if (projectId) {
         setAllCurrentProjectFiles((prev) => [...optimisticFiles, ...prev]);
         projectToUploadFilesMapRef.current.set(projectId, optimisticFiles);
       }
-      svcUploadFiles(files, projectId, tempIdMap)
+      svcUploadFiles(validFiles, projectId, tempIdMap)
         .then((uploaded) => {
           const uploadedFiles = uploaded.user_files || [];
           const tempIdToUploadedFileMap = new Map(
@@ -445,6 +473,7 @@ export function ProjectsProvider({ children }: ProjectsProviderProps) {
       refreshCurrentProjectDetails,
       refreshRecentFiles,
       removeOptimisticFilesByTempIds,
+      settingsContext,
     ]
   );
 

web/src/providers/__tests__/ProjectsContext.test.tsx

@@ -0,0 +1,166 @@
+import React, { PropsWithChildren } from "react";
+import { act, renderHook } from "@testing-library/react";
+import {
+  ProjectsProvider,
+  useProjectsContext,
+} from "@/providers/ProjectsContext";
+import { SettingsContext } from "@/providers/SettingsProvider";
+import { CombinedSettings } from "@/interfaces/settings";
+import type { ProjectFile } from "@/app/app/projects/projectsService";
+
+const mockUploadFiles = jest.fn();
+const mockGetRecentFiles = jest.fn();
+const mockToastWarning = jest.fn();
+
+jest.mock("next/navigation", () => ({
+  useSearchParams: () => ({
+    get: () => null,
+  }),
+}));
+
+jest.mock("@/hooks/appNavigation", () => ({
+  useAppRouter: () => jest.fn(),
+}));
+
+jest.mock("@/lib/hooks/useProjects", () => ({
+  useProjects: () => ({
+    projects: [],
+    refreshProjects: jest.fn().mockResolvedValue([]),
+  }),
+}));
+
+jest.mock("@/hooks/useToast", () => ({
+  toast: {
+    warning: (...args: unknown[]) => mockToastWarning(...args),
+    error: jest.fn(),
+    success: jest.fn(),
+  },
+}));
+
+jest.mock("@/app/app/projects/projectsService", () => {
+  const actual = jest.requireActual("@/app/app/projects/projectsService");
+  return {
+    ...actual,
+    fetchProjects: jest.fn().mockResolvedValue([]),
+    createProject: jest.fn(),
+    uploadFiles: (...args: unknown[]) => mockUploadFiles(...args),
+    getRecentFiles: (...args: unknown[]) => mockGetRecentFiles(...args),
+    getFilesInProject: jest.fn().mockResolvedValue([]),
+    getProject: jest.fn(),
+    getProjectInstructions: jest.fn(),
+    upsertProjectInstructions: jest.fn(),
+    getProjectDetails: jest.fn(),
+    renameProject: jest.fn(),
+    deleteProject: jest.fn(),
+    deleteUserFile: jest.fn(),
+    getUserFileStatuses: jest.fn().mockResolvedValue([]),
+    unlinkFileFromProject: jest.fn(),
+    linkFileToProject: jest.fn(),
+  };
+});
+
+const settingsValue: CombinedSettings = {
+  settings: {
+    user_file_max_upload_size_mb: 1,
+  } as CombinedSettings["settings"],
+  enterpriseSettings: null,
+  customAnalyticsScript: null,
+  webVersion: null,
+  webDomain: null,
+  isSearchModeAvailable: true,
+};
+
+const wrapper = ({ children }: PropsWithChildren) => (
+  <SettingsContext.Provider value={settingsValue}>
+    <ProjectsProvider>{children}</ProjectsProvider>
+  </SettingsContext.Provider>
+);
+
+describe("ProjectsContext beginUpload size precheck", () => {
+  beforeEach(() => {
+    mockUploadFiles.mockReset();
+    mockGetRecentFiles.mockReset();
+    mockToastWarning.mockReset();
+
+    mockUploadFiles.mockResolvedValue({
+      user_files: [],
+      rejected_files: [],
+    });
+    mockGetRecentFiles.mockResolvedValue([]);
+  });
+
+  it("only sends valid files to the upload API when oversized files are present", async () => {
+    const { result } = renderHook(() => useProjectsContext(), { wrapper });
+
+    const valid = new File(["small"], "small.txt", { type: "text/plain" });
+    const oversized = new File([new Uint8Array(2 * 1024 * 1024)], "big.txt", {
+      type: "text/plain",
+    });
+
+    let optimisticFiles: ProjectFile[] = [];
+    await act(async () => {
+      optimisticFiles = await result.current.beginUpload(
+        [valid, oversized],
+        null
+      );
+    });
+
+    expect(mockUploadFiles).toHaveBeenCalledTimes(1);
+    const [uploadedFiles] = mockUploadFiles.mock.calls[0];
+    expect((uploadedFiles as File[]).map((f) => f.name)).toEqual(["small.txt"]);
+    expect(optimisticFiles.map((f) => f.name)).toEqual(["small.txt"]);
+    expect(mockToastWarning).toHaveBeenCalledTimes(1);
+  });
+
+  it("uploads all files when none are oversized", async () => {
+    const { result } = renderHook(() => useProjectsContext(), { wrapper });
+
+    const first = new File(["small"], "first.txt", { type: "text/plain" });
+    const second = new File(["small"], "second.txt", { type: "text/plain" });
+
+    let optimisticFiles: ProjectFile[] = [];
+    await act(async () => {
+      optimisticFiles = await result.current.beginUpload([first, second], null);
+    });
+
+    expect(mockUploadFiles).toHaveBeenCalledTimes(1);
+    const [uploadedFiles] = mockUploadFiles.mock.calls[0];
+    expect((uploadedFiles as File[]).map((f) => f.name)).toEqual([
+      "first.txt",
+      "second.txt",
+    ]);
+    expect(mockToastWarning).not.toHaveBeenCalled();
+    expect(optimisticFiles.map((f) => f.name)).toEqual([
+      "first.txt",
+      "second.txt",
+    ]);
+  });
+
+  it("does not call upload API when all files are oversized", async () => {
+    const { result } = renderHook(() => useProjectsContext(), { wrapper });
+
+    const oversized = new File(
+      [new Uint8Array(2 * 1024 * 1024)],
+      "too-big.txt",
+      { type: "text/plain" }
+    );
+    const onSuccess = jest.fn();
+    const onFailure = jest.fn();
+
+    let optimisticFiles: ProjectFile[] = [];
+    await act(async () => {
+      optimisticFiles = await result.current.beginUpload(
+        [oversized],
+        null,
+        onSuccess,
+        onFailure
+      );
+    });
+
+    expect(mockUploadFiles).not.toHaveBeenCalled();
+    expect(optimisticFiles).toEqual([]);
+    expect(mockToastWarning).toHaveBeenCalledTimes(1);
+    expect(onSuccess).not.toHaveBeenCalled();
+    expect(onFailure).toHaveBeenCalledWith([]);
+  });
+});

web/src/refresh-pages/AppPage.tsx

@@ -1,10 +1,7 @@
 "use client";
 
 import { redirect, useRouter, useSearchParams } from "next/navigation";
-import {
-  personaIncludesRetrieval,
-  getAvailableContextTokens,
-} from "@/app/app/services/lib";
+import { personaIncludesRetrieval } from "@/app/app/services/lib";
 import { useCallback, useEffect, useMemo, useRef, useState } from "react";
 import { toast } from "@/hooks/useToast";
 import { SEARCH_PARAM_NAMES } from "@/app/app/services/searchParams";
@@ -56,10 +53,7 @@ import ChatScrollContainer, {
 } from "@/sections/chat/ChatScrollContainer";
 import ProjectContextPanel from "@/app/app/components/projects/ProjectContextPanel";
 import { useProjectsContext } from "@/providers/ProjectsContext";
-import {
-  getProjectTokenCount,
-  getMaxSelectedDocumentTokens,
-} from "@/app/app/projects/projectsService";
+import { getProjectTokenCount } from "@/app/app/projects/projectsService";
 import ProjectChatSessionList from "@/app/app/components/projects/ProjectChatSessionList";
 import { cn } from "@/lib/utils";
 import Suggestions from "@/sections/Suggestions";
@@ -70,7 +64,6 @@ import * as AppLayouts from "@/layouts/app-layouts";
 import { SvgChevronDown, SvgFileText } from "@opal/icons";
 import { Button } from "@opal/components";
 import Spacer from "@/refresh-components/Spacer";
-import { DEFAULT_CONTEXT_TOKENS } from "@/lib/constants";
 import useAppFocus from "@/hooks/useAppFocus";
 import { useQueryController } from "@/providers/QueryControllerProvider";
 import WelcomeMessage from "@/app/app/components/WelcomeMessage";
@@ -367,18 +360,22 @@ export default function AppPage({ firstMessage }: ChatPageProps) {
   const autoScrollEnabled = user?.preferences?.auto_scroll !== false;
   const isStreaming = currentChatState === "streaming";
 
-  const { onSubmit, stopGenerating, handleMessageSpecificFileUpload } =
-    useChatController({
-      filterManager,
-      llmManager,
-      availableAgents: agents,
-      liveAgent,
-      existingChatSessionId: currentChatSessionId,
-      selectedDocuments,
-      searchParams,
-      resetInputBar,
-      setSelectedAgentFromId,
-    });
+  const {
+    onSubmit,
+    stopGenerating,
+    handleMessageSpecificFileUpload,
+    availableContextTokens,
+  } = useChatController({
+    filterManager,
+    llmManager,
+    availableAgents: agents,
+    liveAgent,
+    existingChatSessionId: currentChatSessionId,
+    selectedDocuments,
+    searchParams,
+    resetInputBar,
+    setSelectedAgentFromId,
+  });
 
   const { onMessageSelection, currentSessionFileTokenCount } =
     useChatSessionController({
@@ -595,43 +592,6 @@ export default function AppPage({ firstMessage }: ChatPageProps) {
     };
   }, [currentChatSessionId, currentProjectId, currentProjectDetails?.files]);
 
-  // Available context tokens source of truth:
-  // - If a chat session exists, fetch from session API (dynamic per session/model)
-  // - If no session, derive from the default/current persona's max document tokens
-  const [availableContextTokens, setAvailableContextTokens] = useState<number>(
-    DEFAULT_CONTEXT_TOKENS * 0.5
-  );
-  useEffect(() => {
-    let cancelled = false;
-    async function run() {
-      try {
-        if (currentChatSessionId) {
-          const available =
-            await getAvailableContextTokens(currentChatSessionId);
-          const capped_context_tokens =
-            (available ?? DEFAULT_CONTEXT_TOKENS) * 0.5;
-          if (!cancelled) setAvailableContextTokens(capped_context_tokens);
-        } else {
-          const personaId = (selectedAgent || liveAgent)?.id;
-          if (personaId !== undefined && personaId !== null) {
-            const maxTokens = await getMaxSelectedDocumentTokens(personaId);
-            const capped_context_tokens =
-              (maxTokens ?? DEFAULT_CONTEXT_TOKENS) * 0.5;
-            if (!cancelled) setAvailableContextTokens(capped_context_tokens);
-          } else if (!cancelled) {
-            setAvailableContextTokens(DEFAULT_CONTEXT_TOKENS * 0.5);
-          }
-        }
-      } catch (e) {
-        if (!cancelled) setAvailableContextTokens(DEFAULT_CONTEXT_TOKENS * 0.5);
-      }
-    }
-    run();
-    return () => {
-      cancelled = true;
-    };
-  }, [currentChatSessionId, selectedAgent?.id, liveAgent?.id]);
-
   // handle error case where no assistants are available
   // Only show this after agents have loaded to prevent flash during initial load
   if (noAgents && !isLoadingAgents) {

web/tests/e2e/admin/llm_provider_setup.spec.ts

@@ -21,10 +21,47 @@ type DefaultModelInfo = {
   model_name: string;
 } | null;
 
+type ProviderModelConfig = {
+  name: string;
+  is_visible: boolean;
+};
+
 function uniqueName(prefix: string): string {
   return `${prefix}-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
 }
 
+function normalizeAlphaNum(input: string): string {
+  return input.toLowerCase().replace(/[^a-z0-9]/g, "");
+}
+
+function modelTokenVariants(modelName: string): string[][] {
+  return modelName
+    .toLowerCase()
+    .split(/[^a-z0-9]+/)
+    .filter((token) => token.length > 0)
+    .map((token) => {
+      // Display names may shorten long numeric segments to suffixes.
+      if (/^\d+$/.test(token) && token.length > 5) {
+        return [token, token.slice(-5)];
+      }
+      return [token];
+    });
+}
+
+function textMatchesModel(modelName: string, candidateText: string): boolean {
+  const normalizedCandidate = normalizeAlphaNum(candidateText);
+  if (!normalizedCandidate) {
+    return false;
+  }
+
+  const tokenVariants = modelTokenVariants(modelName);
+  return tokenVariants.every((variants) =>
+    variants.some((variant) =>
+      normalizedCandidate.includes(normalizeAlphaNum(variant))
+    )
+  );
+}
+
 async function getAdminLLMProviderResponse(page: Page) {
   const response = await page.request.get(`${BASE_URL}/api/admin/llm/provider`);
   expect(response.ok()).toBeTruthy();
@@ -50,6 +87,18 @@ async function createPublicProvider(
   providerName: string,
   modelName: string = "gpt-4o"
 ): Promise<number> {
+  return createPublicProviderWithModels(page, providerName, [
+    { name: modelName, is_visible: true },
+  ]);
+}
+
+async function createPublicProviderWithModels(
+  page: Page,
+  providerName: string,
+  modelConfigurations: ProviderModelConfig[]
+): Promise<number> {
+  expect(modelConfigurations.length).toBeGreaterThan(0);
+
   const response = await page.request.put(
     `${BASE_URL}/api/admin/llm/provider?is_creation=true`,
     {
@@ -60,7 +109,7 @@ async function createPublicProvider(
         is_public: true,
         groups: [],
         personas: [],
-        model_configurations: [{ name: modelName, is_visible: true }],
+        model_configurations: modelConfigurations,
       },
     }
   );
@@ -69,6 +118,86 @@ async function createPublicProvider(
   return data.id;
 }
 
+async function navigateToAdminLlmPageFromChat(page: Page): Promise<void> {
+  await page.goto(LLM_SETUP_URL);
+  await page.waitForURL("**/admin/configuration/llm**");
+  await expect(page.getByLabel("admin-page-title")).toHaveText(
+    /^Language Models/
+  );
+}
+
+async function exitAdminToChat(page: Page): Promise<void> {
+  await page.goto("/app");
+  await page.waitForURL("**/app**");
+  await page
+    .locator("#onyx-chat-input-textarea")
+    .waitFor({ state: "visible", timeout: 15000 });
+}
+
+async function isModelVisibleInChatProviders(
+  page: Page,
+  modelName: string
+): Promise<boolean> {
+  const response = await page.request.get(`${BASE_URL}/api/llm/provider`);
+  expect(response.ok()).toBeTruthy();
+
+  const data = (await response.json()) as {
+    providers: {
+      model_configurations: { name: string; is_visible: boolean }[];
+    }[];
+  };
+
+  return data.providers.some((provider) =>
+    provider.model_configurations.some(
+      (model) => model.name === modelName && model.is_visible
+    )
+  );
+}
+
+async function expectModelVisibilityInChatProviders(
+  page: Page,
+  modelName: string,
+  expectedVisible: boolean
+): Promise<void> {
+  await expect
+    .poll(() => isModelVisibleInChatProviders(page, modelName), {
+      timeout: 30000,
+    })
+    .toBe(expectedVisible);
+}
+
+async function getModelCountInChatSelector(
+  page: Page,
+  modelName: string
+): Promise<number> {
+  const dialog = page.locator('[role="dialog"]').first();
+
+  // When used in expect.poll retries, a previous attempt may leave the
+  // popover open. Ensure a clean state before toggling it.
+  if (await dialog.isVisible()) {
+    await page.keyboard.press("Escape");
+    await dialog.waitFor({ state: "hidden", timeout: 5000 });
+  }
+
+  await page.getByTestId("AppInputBar/llm-popover-trigger").click();
+  await dialog.waitFor({ state: "visible", timeout: 10000 });
+
+  await dialog.getByPlaceholder("Search models...").fill(modelName);
+  const optionButtons = dialog.getByRole("button");
+  const optionTexts = await optionButtons.allTextContents();
+  const uniqueOptionTexts = Array.from(
+    new Set(optionTexts.map((text) => text.trim()))
+  );
+  const count = uniqueOptionTexts.filter((text) =>
+    textMatchesModel(modelName, text)
+  ).length;
+
+  await page.keyboard.press("Escape");
+  await dialog.waitFor({ state: "hidden", timeout: 10000 });
+
+  return count;
+}
+
 async function getProviderByName(
   page: Page,
   providerName: string
@@ -272,4 +401,132 @@ test.describe("LLM Provider Setup @exclusive", () => {
       }
     }
   });
+
+  test("adding a hidden model on an existing provider shows it in chat after one save", async ({
+    page,
+  }) => {
+    await page.route("**/api/admin/llm/test", async (route) => {
+      await route.fulfill({
+        status: 200,
+        contentType: "application/json",
+        body: JSON.stringify({ success: true }),
+      });
+    });
+
+    const providerName = uniqueName("PW Provider Add Model");
+    const ts = Date.now();
+    const alwaysVisibleModel = `pw-visible-${ts}-base`;
+    const modelToEnable = `pw-hidden-${ts}-to-enable`;
+
+    const providerId = await createPublicProviderWithModels(
+      page,
+      providerName,
+      [
+        { name: alwaysVisibleModel, is_visible: true },
+        { name: modelToEnable, is_visible: false },
+      ]
+    );
+    providersToCleanup.push(providerId);
+    await expectModelVisibilityInChatProviders(page, modelToEnable, false);
+
+    await page.goto("/app");
+    await page.waitForLoadState("networkidle");
+    await page
+      .locator("#onyx-chat-input-textarea")
+      .waitFor({ state: "visible", timeout: 15000 });
+
+    await expect
+      .poll(() => getModelCountInChatSelector(page, modelToEnable), {
+        timeout: 15000,
+      })
+      .toBe(0);
+
+    await navigateToAdminLlmPageFromChat(page);
+
+    const editModal = await openProviderEditModal(page, providerName);
+    await editModal.getByText(modelToEnable, { exact: true }).click();
+
+    const updateButton = editModal.getByRole("button", { name: "Update" });
+    const providerUpdateResponsePromise = page.waitForResponse(
+      (response) =>
+        response.url().includes("/api/admin/llm/provider") &&
+        response.request().method() === "PUT"
+    );
+    await expect(updateButton).toBeEnabled({ timeout: 10000 });
+    await updateButton.click();
+    await providerUpdateResponsePromise;
+    await expect(editModal).not.toBeVisible({ timeout: 30000 });
+    await expectModelVisibilityInChatProviders(page, modelToEnable, true);
+
+    await exitAdminToChat(page);
+    await expect
+      .poll(() => getModelCountInChatSelector(page, modelToEnable), {
+        timeout: 15000,
+      })
+      .toBe(1);
+  });
+
+  test("removing a visible model on an existing provider hides it in chat after one save", async ({
+    page,
+  }) => {
+    await page.route("**/api/admin/llm/test", async (route) => {
+      await route.fulfill({
+        status: 200,
+        contentType: "application/json",
+        body: JSON.stringify({ success: true }),
+      });
+    });
+
+    const providerName = uniqueName("PW Provider Remove Model");
+    const ts = Date.now();
+    const alwaysVisibleModel = `pw-visible-${ts}-base`;
+    const modelToDisable = `pw-visible-${ts}-to-disable`;
+
+    const providerId = await createPublicProviderWithModels(
+      page,
+      providerName,
+      [
+        { name: alwaysVisibleModel, is_visible: true },
+        { name: modelToDisable, is_visible: true },
+      ]
+    );
+    providersToCleanup.push(providerId);
+    await expectModelVisibilityInChatProviders(page, modelToDisable, true);
+
+    await page.goto("/app");
+    await page.waitForLoadState("networkidle");
+    await page
+      .locator("#onyx-chat-input-textarea")
+      .waitFor({ state: "visible", timeout: 15000 });
+
+    await expect
+      .poll(() => getModelCountInChatSelector(page, modelToDisable), {
+        timeout: 15000,
+      })
+      .toBe(1);
+
+    await navigateToAdminLlmPageFromChat(page);
+
+    const editModal = await openProviderEditModal(page, providerName);
+    await editModal.getByText(modelToDisable, { exact: true }).click();
+
+    const updateButton = editModal.getByRole("button", { name: "Update" });
+    const providerUpdateResponsePromise = page.waitForResponse(
+      (response) =>
+        response.url().includes("/api/admin/llm/provider") &&
+        response.request().method() === "PUT"
+    );
+    await expect(updateButton).toBeEnabled({ timeout: 10000 });
+    await updateButton.click();
+    await providerUpdateResponsePromise;
+    await expect(editModal).not.toBeVisible({ timeout: 30000 });
+    await expectModelVisibilityInChatProviders(page, modelToDisable, false);
+
+    await exitAdminToChat(page);
+    await expect
+      .poll(() => getModelCountInChatSelector(page, modelToDisable), {
+        timeout: 15000,
+      })
+      .toBe(0);
+  });
 });