Batch write updates to db. Triggered every minute. NOTE: This job now skips tag spend updates, which are handled by a separate sche

A minimal-but-valid dict for ``LiteLLMBatch(**response.model_dump())``.

Normalise a passthrough route to a bare /v1/... path for map lookup. Examples: /openai_passthrough/v1/files -> /v1/files /open

Evaluate content against Purview DLP policies. Args: user_id: Entra object ID. text: Content to evaluate.

Check if the team key is allocating guaranteed throughput limits. If so, raise an error if we're overallocating. Only runs check if tpm_limi

Clear all names from litellm globals.

Common checks for 'get_available_deployment' across sync + async call. If 'healthy_deployments' returned is None, this means the use

Helper function to convert pre_call_hook response back to kwargs for MCP usage. Supports: - modified_arguments: Override too

Return True the first time this handler is asked to emit a span for the given (handler, scope) on this kwargs; False on repeats. Used

Expand ``aliases`` lists in model cost entries into top-level entries. Each alias gets a reference to the **same** dict object as the canoni

Build a minimal stand-in for ``UserAPIKeyAuth`` with just the fields consulted by ``_check_summary_model_access`` and ``_check_summary_model_b

Extract the value set for a specific attribute name, or None.

Internal debugging helper function Masks the headers of the request sent from LiteLLM Args: masked_length: Optional length for

Helper to return what the "metadata" field should be called in the request data For all /thread or /assistant endpoints we need to call this

Hash the token if it's a string and starts with "sk-" Else return the token as is

Check if the error message indicates an idempotent operation error. Idempotent errors (like "column already exists") mean the migrat

Returns True if the call type is a streaming request. Returns True if: - if "stream=True" in kwargs (litellm chat completion, litell

Create a mock _get_session_iterator that yields a session returning json_response.

Build auth and extra headers for a server.

Extract and prepare metadata from request data for PANW API call. Supported metadata fields (from request.metadata): - profi

Return a prisma_client whose find_many honors the provider scope pushed into the ``where`` clause, mirroring how Postgres would filter rows.

An object that mimics the OpenAI SDK Batch: only ``.model_dump()`` is used.

Set up mocks for bulk_update_team_keys; returns mock_prisma.

Build a StandardLoggingGuardrailInformation entry the way ``add_standard_logging_guardrail_information_to_request_data`` does.

Transform OpenAI-style input to Nova format. Only handles OpenAI params (dimensions). All other Nova-specific params should

Helper function to check if user is an org admin for all of the passed organizations. Checks both: - `organization_id` (singular string)

Verify that only the requested name is in globals, not the others.

Add a failure callback to `litellm.failure_callback`. Auto-routes async callbacks to litellm._async_failure_callback.

Key - the key in cache Value - float - the value you want to increment by Refresh_ttl - bool - if True, resets the Redis TT

For usage-based-routing-v2, enables running rpm checks before the call is made, inside the semaphore. -> makes the calls concurrency

Authenticate a user and generate an API key for UI access. This function handles two login scenarios: 1. Admin login using UI_USERNAME a

Call a tool with the given name and arguments Args: server_name: Server name name: Tool name arg

Return True iff the caller may read/modify a managed resource. The resource's ``created_by`` and ``team_id`` fields must be non-None to match

Check if a specific permission is allowed for a given vector store index. Args: index_name: The name of the vector store index

Cleanup Redis cache before each test

OpenAI message with a tool result looks like: { "tool_call_id": "tool_1", "role": "tool", "content": "function result

Create model_call_details dict with standard_logging_object.

Create a standardized OpenAI-compatible model object. When include_metadata is true, attaches the model's configured fallbacks (resolved

Delete a field from nested data using JSONPath notation. Custom implementation - no external dependencies. Supports: - "field" - to

Decode a cache entry to ``model_type``. - ``None`` → ``None`` - Already an instance of ``model_type`` (including subclasses)

Rewrite an ``mcp_tool_permissions`` dict keyed by id/name/alias so every key is a concrete server_id where possible. Tool lists from

Creates appropriate wrapper functions for different API call types. Returns: - A synchronous function for synchronous ca

Log failed db read/writes Currently only logs exceptions to sentry

If a model has a team_id Only use if request is from that team

Retrieves the Arize Phoenix configuration based on environment variables. Returns:

Get custom labels from metadata

Returns `extra_body` to be added to the request body for the Guardrail API call Use this to pass dynamic params to the guardrail API

Determine MIME type from file path extension. Falls back to mimetypes.guess_type, then to 'application/octet-stream'.

v2 vector store config, use this for new vector store integrations

Clean and filter the metadata dictionary to include only the specified keys in StandardLoggingMetadata. Args: metadata (