Walk the model_config precedence chain in team/project metadata. If a matching credential is found, set api_base/api_key/api_version on data

OpenAI tools looks like: tools = [ { "type": "function", "function": { "name": "get_current_w

Build a managed container ID with provider and model info encoded. Format: cntr_{base64("litellm:custom_llm_provider:{provider};model_id:{mod

Build filter conditions for key listing. Visibility rules: - Users always see their own keys (user_id match) - Team admins see ALL keys f

Build the User-Agent string for Databricks API calls. If a custom user agent is provided, the partner name (part before /) i

Creates a span for Guardrail, if any guardrail information is present in standard_logging_object

Utility to create a mocked async httpx client for the given method.

Get the API endpoint for a given model.

Build a minimal mock ResponsesAPIResponse.

Build a router model_list from a compact spec. spec entry shape: { "model_name": str, "quality_tier": Optional[int],

Create a mock team object with the given MCP permissions.

Resolve the target Phoenix/Arize project for this request. Proxy priority: ``user_api_key_auth_metadata.phoenix_project_name_overrid

Transforms ResponseAPIUsage or ImageUsage to a Usage object. Both have the same spec with input_tokens, output_tokens, and i

Try to transform Vertex AI batch output to OpenAI format. If conversion fails at any point, return the original content as-is.

Build a mocked Repello analyze response with the given verdict.

The four required attributes on the SERVER span must all be set.

Try calling the function_with_retries If it fails after num_retries, fall back to another model group

Check if route is a passthrough route. Supports both exact match and prefix match.

Convert a file-type document dict to a document_url-type document dict with an inline base64 data URI. Accepts document dicts like:

Validate and configure the Databricks environment. Authentication priority: 1. OAuth M2M (DATABRICKS_CLIENT_ID + DATABRICKS_

Decode provider and model_id from encoded video_id.

Encode a file/batch ID with model routing information. Format: <prefix><base64(litellm:<original_id>;model,<model_name>)> The result pre

Add a coroutine to the logging queue. Hot path: never blocks, aggressively clears queue if full.

Execute pipeline steps sequentially with conditional actions. Args: steps: Ordered list of pipeline steps mo

Expand a permission list of server_ids/names/aliases into concrete server_ids against the current region's config + DB registry union

Read config from the environment, falling back to defaults. An invalid value warns and uses the default rather than failing startup.

Resolve Anthropic credentials and return the appropriate auth header dict. Checks ANTHROPIC_API_KEY first (-> x-api-key or Bearer depending o

Get the budget reset time based on the configured timezone. Falls back to UTC if not specified.

Get API credentials and provider info from a model name in model_list. Useful for passthrough endpoints (files, batches, etc.) that n

Get the model rpm limit for a given api key. Priority order (returns first found): 1. Key metadata (model_rpm_limit) 2. Key model_ma

Returns the supported openai params for a given model + provider Example: ``` get_supported_openai_params(model="anthropic.claude-3"

Map OCR parameters to provider-specific parameters.

Serialize a Pydantic model to a dictionary while preserving specific fields even if they are None. Fields listed in _PRESERVED_NONE_FIEL

Run pre-call checks and guardrail hooks for an MCP tool call. Returns a dict that may contain: - "arguments": hook-modified

A call that should have succeeded but didn't is a hard failure, never a skip: if the proxy can't make a call it's expected to, the test must fail.

Optionally drop end-user ids that don't resolve to a known DB row. Default: pass-through. LiteLLM's documented pattern is that the `user` fie

Resolve ``(model, llm_provider)`` for a request being rejected by an internal proxy-side rate-limit hook. These hooks fire from ``async_

Open a session, run the provided coroutine, and clean up.

Only runs if router.slack_alerting_logger is set Sends a Slack / MS Teams alert for the LLM API call failure. Only if router.slack_alerting_l

Determine if message logging should be redacted. Priority order: 1. Dynamic parameter (turn_off_message_logging in request) 2. Heade

Determine if polling via cache should be used for a request. Args: background_mode: Whether background=true was set in the request

Log a spend-tracking error, with the traceback gated on the env var. By default this behaves like ``verbose_proxy_logger.exception`` — the ac

Transform the request for Manus API. Manus requires: - `task_mode: "agent"` in the request body - `agent_profile` ex

Return a usable end-user identifier string, or None if the value isn't one. Always drops non-string structured values (dict/list/tuple/set) becau

Create the standard logging object for Anthropic passthrough handles streaming and non-streaming responses

Decode the responses_api_response_id Returns: DecodedResponseId: Structured tuple with custom_llm_provider, model_id, an

Invalidate the case-insensitive lookup map for model_cost. Call this whenever litellm.model_cost is modified to ensure the map is rebuilt. Al

Check if user is an org admin for the team's organization. Returns True if: - The team belongs to an organization, AND - The user ha

Process a single chunk of data from the stream

Returns: - ``UserAPIKeyAuth``: a resolved virtual key (cache hit or DB hit). The caller may use this directly; JWT policy has been

Run a reconnect cycle with a single overall timeout budget. Uses the _engine_confirmed_dead flag (set by waitpid thread / pidfd / po

Sign a request for Bedrock or Sagemaker Returns: Tuple[dict, Optional[str]]: A tuple containing the headers and the json

Add server name prefix to any MCP resource name.