Return the row in the db

DualCache is a cache implementation that updates both Redis and an in-memory cache simultaneously. When data is updated or inserted, it is wr

LiteLLM Params without 'model' arg (used across completion / assistants api)

Logging/Custom Handlers for proxy. Implemented mainly to: - log successful/failed db read/writes - support the max parallel request

Abstraction class on top of config loading/updating logic. Gives us one place to control all config updating logic.

Reference: https://docs.anthropic.com/claude/reference/messages_post to pass metadata to anthropic, it's {"user_id": "any-relevant-informati

Reference - https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_Converse.html #2 - https://docs.aws.amazon.com/bedrock/latest

A class to define the roles and permissions for a LiteLLM Proxy w/ JWT Auth. Attributes: - admin_jwt_scope: The JWT scope required for p

LiteLLM Params with 'model' requirement - used for completions

- treat the sub id passed in as the user id - return an error if id making request doesn't exist in proxy user table - track spend agains

Complexity router that classifies requests and routes to appropriate models. By default, handles requests in <1ms with zero external API cal

ModelResponse represents a response from the API containing model information.

Reference: https://cloud.google.com/vertex-ai/docs/generative-ai/chat/test-chat-prompts Reference: https://cloud.google.com/vertex-ai/generat

Represents a LiteLLM_ObjectPermissionTable record

Content filter guardrail that detects sensitive information using: - Prebuilt regex patterns (SSN, credit cards, API keys, etc.) - Custom

LiteLLM BaseConfig implementation for OCI Generative AI chat.

POST /model/new litellm_params: `model` is the only required field; `api_key` et al may be an `os.environ/FOO` reference the proxy resolves at cal

CustomLogger that intercepts WebSearch tool calls for models that don't natively support web search. Implements agentic loop: 1. Det

Represents the url or the content of an image generated by the OpenAI API. Attributes: b64_json: The base64-encoded JSON of the generate

Call Claude model family in the /v1/messages API spec Supports anthropic_beta parameter for beta features.

A user's OAuth credential: the bearer value, when it expires, and how to refresh it. ``expires_at`` is epoch seconds (``None`` means no known exp

Base for header models. Subclasses may alias to hyphenated header names (e.g. ``x-litellm-api-key``); serialization uses by_alias.

Represents user-controllable params for a LiteLLM_BudgetTable record. Budget-write paths use `model_fields.keys()` on this class as an allowlist

Represents a LiteLLM_MCPServerTable record

Repository for user database operations.

- first chunk return 'message_start' - content block must be started and stopped - finish_reason must map exactly to anthropic reason, el

DualCache wrapper for UserAPIKeyAuth-like payloads. Stores a Redis-safe JSON payload in BOTH in-memory and Redis to avoid "memory return

Repository for team database operations.

Produces the one `httpx.Auth` for a `(subject, upstream)` pair, per declared mode. Collaborators (the per-mode credential stores and token fetche

Empty body/query for routes that take none.

CustomLogger that implements transparent code-interpreter execution loops. Flow: 1. Replace the native ``code_interpreter`` tool with a