Enqueue an update.

Sync: List all agents on the provider side.

Async version of _ensure_access_token

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

Build a fake acompletion response with the given message content.

Reset all callbacks to an empty list Note: this is an internal function and should be used sparingly.

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

Retrieves the cached result for the given arguments. Args: *args: args to litellm.completion() or embedding()

Get a dict for the maximum tokens (context window), input_cost_per_token, output_cost_per_token for a given model. Parameters: - model

Return JSON response data.

Validate a user-supplied URL and rewrite it to connect to a validated IP. Resolves the hostname, checks all resolved IPs against blocked net

Classify a prompt by complexity, using the LLM classifier when configured. Falls back to the local heuristic scorer if classifier_ty

Apply ``compact_20260112``; return a ``PolyfillResult``. See module docstring for the algorithm. Errors are best-effort: when the summary cal

Classify a prompt by complexity, discarding which rule decided the tier. Kept for callers that only need the tier and score; `_score_and_clas

Pass through endpoint handler, makes the httpx request for pass-through endpoints and ensures logging hooks are called Args: request

Patch the endpoint module's _require_prisma to return our fake.

3 jobs: - Adds selected deployment, model_info and api_base to kwargs["metadata"] (used for logging) - Adds default litellm p

format messages for anthropic 1. Anthropic supports roles like "user" and "assistant" (system prompt sent separately) 2. The first messag

Asynchronously set cache using run_in_executor to avoid blocking the event loop. Compatible with Python 3.8+.

Main authentication and authorization builder

Extract client IP from a FastAPI request for MCP access control. Security: Only trusts X-Forwarded-For if: 1. use_x_forwarde

Return the prefix for a server. When the short-prefix mode is enabled (``LITELLM_USE_SHORT_MCP_TOOL_PREFIX``) a three-character base62 ID is

Translate messages to anthropic format.

Mock pre-call hook that raises BlockedPiiEntityError

Get the base URL for the request, considering X-Forwarded-* headers. Resolution order: ``PROXY_BASE_URL`` env var, then X-Forwarded-* when

Validate ``candidate`` and recover the identity and inner grant. Never raises for bad input: every invalid, expired, tampered, or undecryptable

Return a UUID4 using the selected backend.

MakeRequest is a helper function to make HTTP requests

Asynchronously executes a litellm.completion() call for any of litellm supported llms (example gpt-4, gpt-3.5-turbo, claude-2, command-nightly)

Async get cache implementation. Used for embedding calls in async wrapper