BerriAI

LiteLLM Audit Logging & Multi-Provider Evidence

Python multi-vendor proxy with one evidence hook

LiteLLM is the dominant Python routing layer — an OpenAI-compatible proxy across 100+ providers. Sigigo middleware logs every completion with routed model metadata and commitments at the proxy boundary.

Where Sigigo sits

At the LiteLLM completion boundary — callback hooks or a thin proxy wrapper around litellm.completion and async variants. One integration point covers every upstream provider your proxy routes to.

Why Python teams use LiteLLM

ML and data science teams standardize on LiteLLM instead of per-vendor SDKs:

  • Single OpenAI-compatible interface for Bedrock, Vertex, Anthropic, and more
  • Centralized routing, budgeting, and fallback in one service
  • Common in FastAPI backends and batch inference pipelines
  • When Article 12 applies, automatic recording should attach at the proxy — not only at individual provider SDKs

Integration shape

@sigigo/integrations-litellm — success/failure callbacks mapped to ai_act events.

  • litellm.callbacks → SigigoCallbackHandler
  • Routed model id and provider on every event
  • Composable with LangChain when both are in the same stack
  • EU preset enables commitment mode by default

Evidence events

Canonical event types emitted through this integration path:

  • ai_act.inference_started / inference_completed
  • ai_act.input_received (commitment)
  • ai_act.output_generated
  • gpai_model_id from LiteLLM routing metadata
  • ai_act.action_executed (tool / function calls)

Related

Integration availability may vary by tenant, region, and deployment model. This page describes product architecture — not a guarantee of feature availability. See compliance disclaimers and contact us for early access.