Helper function to get the bedrock provider from the model handles 4 scenarios: 1. model=invoke/anthropic.claude-3-5-sonnet-

Get the complete A2A agent endpoint URL. A2A agents use JSON-RPC 2.0 at the base URL, not specific paths. The method (messag

Get custom labels from metadata

Get the `custom_llm_provider` from the request body Safely reads the request body

Common function to get daily activity for any entity type. ``resolve_entity_metadata`` lets a caller resolve entity metadata from the rows ac

Fetch the content of a file from the GitLab repository at the given ref (tag, branch, or commit SHA). If `ref` is None, uses self.ref

Returns all possible model names for the router, including models defined via model_group_alias. If a team_id is provided, only depl

Concatenate text from every chat message (all roles) for pre-call DLP. Evaluates the same payload the model receives, not only the trailing u

Get provider-specific Evals API configuration Args: provider: The LLM provider Returns: Provider-sp

Filter parameters to only include those defined in AnthropicMessagesRequestOptionalParams. Args: params: Dictionary of p

Unified SSL configuration function that handles ssl_context and ssl_verify logic. SSL Configuration Priority: 1. If ssl_verify is provid

Retrieve usage data from LiteLLM daily user spend table.

Health check endpoint for the prisma client

🚨 USE `/health/liveliness` to health check the proxy 🚨 See more 👉 https://docs.litellm.ai/docs/proxy/health Check the health of all th

Initialize the standard callback dynamic params from the kwargs checks supported request callback params in kwargs and sets the correspondin

Check if a client IP is from an internal/private network. Handles X-Forwarded-For comma chains (takes leftmost = original client).

Whether this worker has begun graceful shutdown.

Check if tool name has a known MCP server prefix. When ``known_server_prefixes`` is provided the function verifies that the substring be

Map parameters for XAI Responses API. Handles XAI-specific transformations: 1. Drops 'instructions' parameter (not supported

Merge outbound HTTP headers for A2A agent calls. Merge rules: - Start with ``dynamic_headers`` (values extracted from the incoming client req

Handle OpenAI passthrough logging with cost tracking for chat completions, image generation, image editing, and responses API.

Parse tool call arguments from a JSON string. When the JSON is malformed (e.g. truncated by the model), this function attempts a lightwe

Patch a user according to SCIM v2 protocol

Thompson-sample across eligible models. Stateless per-turn.

Register a plugin in the LiteLLM marketplace. LiteLLM acts as a registry/discovery layer. Plugins are hosted on GitHub/GitLab/Bitbucket.

Resolve ``candidate`` only when it is an existing supported image file.

Validate, wire the environment, and hand off to the agent. On success this replaces the current process and never returns. Raises AgentRunErr

Loops through all the fallback model groups and calls kwargs["original_function"] with the arguments and keyword arguments provided. If the

Safe Deep Copy The LiteLLM request may contain objects that cannot be pickled/deep-copied (e.g., tracing spans, locks, clients). Th

Read-only copy of the per-session VCR stats. Used by the summary.

Setup guardrail with all GDPR patterns (mimics the policy template)

Return the unprefixed name plus the server name used as prefix.

Accept the authorization code from client and exchange it for OAuth token. Supports PKCE flow by forwarding code_verifier to upstream provide

Transform the optional params to the format expected by the Vertex AI API. For example, "aspect_ratio" is transformed to "aspectRatio

Transform an error message to Anthropic format. - If already in Anthropic format: passthrough unchanged - Otherwise: extract

Truncate tool names that exceed OpenAI's 64-character limit. Uses format: {55-char-prefix}_{8-char-hash} to avoid collisions when multip

Poll the in-flight request counter until it reaches the drain target or ``timeout`` seconds elapse. Args: timeou

Helper to iterate over a streaming response. Catches errors for fallbacks using the router's fallback system

Extract citations and search_results from Perplexity API response and add them as ChatCompletionAnnotation objects to the message.

Internal method to get from the cache. Handles different call types (embeddings, chat/completions, text_completion, transcription)

Resolve stdio env values, supporting header-driven placeholders.

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

Check if a tool is allowed based on the configured rules Args: tool_name: Name of the tool to check tool_typ

Clean up pidfd reader, waitpid thread ref, or stop polling and reset state.