Get the labels for a metric, filtered if configured. The result is cached on the instance so the label set used to construct

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

Check if a callback is disabled via the x-litellm-disable-callbacks header or via `litellm_disabled_callbacks` in standard_callback_dynam

Safely get Vertex AI project without mutating the litellm_params dict. Unlike get_vertex_ai_project(), this does NOT pop values from

Encode a value for DualCache / Redis (``json.dumps``-safe). If ``model_type`` is set, the payload is validated with that model, then

Returns True if the guardrail should be run on the event_type

Transform LiteLLM data to CBF format, dropping records with zero successful_requests or invalid CZRNs.

No transformation is needed for Anthropic messages This takes in a request in the Anthropic /v1/messages API spec -> transforms it

Translate a full Anthropic /v1/messages request dict to litellm.responses() / litellm.aresponses() kwargs.

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

Checks if token can call a given model Args: - model: str - llm_router: Optional[Router] - models: List[str]

Check if the model has guardrails defined and merge them with existing guardrails in the request data. Args: data: The request data

Internal function. Returns the token and url for the call. Handles logic if it's google ai studio vs. vertex ai. Returns

Helper method to fetch tools from MCP servers based on server filtering criteria. Args: user_api_key_auth: User authenti

Build a stub httpx-like response.

Strip echoed request input from provider error strings. Provider validation errors (e.g. OpenAI ``RateLimitError`` carrying 178 pydantic

Validate permissions and constraints for key update.

Example Usage: ``` from litellm import Router client = Router(model_list = [ { "model_na

Async implementation of 'get_available_deployments'. Allows all cache calls to be made async => 10x perf impact (8rps -> 100 rps).

Check if user has permission to modify (delete/regenerate) a verification token. Rules: - Proxy admin can modify any key - Internal

Create a tool function for an OpenAPI operation. This function creates an async tool function that can be called with keyword arguments. Para

Get the base model from the given model name. Handle model names like: - "us.meta.llama3-2-11b-instruct-v1:0" -> "meta.llama3-2-11b-inst

Get all custom loggers that are instances of the given class type

if 'model_name' is none, returns all. Returns list of model id's. Optimized with O(1) or O(k) index lookup when model_name

Get a proxy hook from the proxy_hook_mapping

Parse a single GENERIC SSE chunk into a LiteLLM ModelResponseStream.

Build the LiteLLM-fronted agent card. Args: upstream_card: Card returned by the upstream agent's well-known endpoint. Ma

Mint a fresh managed ID for a given raw provider ID.

The exact `data` dict the processor was constructed with.

Call from the post-yield section of an autouse fixture per test.

Resolve a single-secret search API key, falling back to a server-managed secret only when the request targets a trusted host.

Safely parse a JSON string. If parsing fails, return the default value (None by default).

Check if the given model supports reasoning and return a boolean value.

Map a v1 server onto a ServerSpec for a migrated mode, or None to defer to v1. BYOK is the per-user source of the ``api_key`` mode; its scheme ri

Authenticate with AWS Role

Create the litellm_request span then close the proxy span.

Create a mock team object with the given MCP permissions.

Return server.static_headers with ``${NAME}`` interpolated. Globals come from ``server.env_vars`` entries with ``scope=="global"``. P

Run ``common_checks`` once at the ``user_api_key_auth`` wrapper boundary, regardless of which ``_user_api_key_auth_builder`` path returned. Th

Recursively sanitize request body to prevent logging large base64 strings or other large values. Truncates strings longer than MAX_STRING_LEN

Serialize a streaming chunk to a JSON string for WebSocket transmission.

Bedrock doesn't support tool calling without `tools=` param specified.

Passes the entire stream to the guardrail This is useful for guardrails that need to see the entire response, such as PII masking.

Extract model name from an encoded file/batch ID. Handles IDs that start with "file-" or "batch_" prefix.

Parameters: - duration: - "<number>s" - seconds - "<number>m" - minutes - "<number>h" - hours - "<number>d" -