LangChain

LangChain & LangGraph Audit Logging for EU AI Act

Agent and RAG traces without duplicate emits

LangChain and LangGraph callbacks capture retrieval, tool calls, and multi-step agent runs — mapped to the same ai_act schema as provider wrappers.

Where Sigigo sits

Sigigo attaches via LangChain callbacks and LangGraph middleware at orchestration boundaries — retrieval, model calls, and tool execution — without requiring manual emit calls in every node.

What you can prove

End-to-end agent evidence for auditors and incident response:

  • Which documents were retrieved and when (commitment over content)
  • Tool calls bound to the same session_id as the originating turn
  • Multi-step LangGraph flows with step-level linkage
  • No duplicate events when stacked with a provider wrapper

Integration shape

@sigigo/integrations-langchain — shared core proof library, thin callback adapter.

  • Retriever callback → ai_act.context_retrieved
  • Tool middleware → ai_act.action_executed
  • Composable with OpenAI / Anthropic wrappers on the same session

Evidence events

Canonical event types emitted through this integration path:

  • ai_act.inference_started / inference_completed
  • ai_act.context_retrieved (RAG steps)
  • ai_act.input_received (commitment)
  • ai_act.action_executed (tool calls)
  • ai_act.output_generated

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.