Google

Google Gemini API Audit Logging & EU AI Act Evidence

Direct Gemini API evidence at the SDK boundary

Google's Gen AI SDK powers a large share of direct model integrations. Sigigo wraps generateContent and streaming calls so Gemini deployments get the same tamper-evident ai_act chain as OpenAI and Claude.

Where Sigigo sits

At the @google/genai client boundary — before requests reach Google and after responses return. The wrapper maps Gemini-specific metadata (model, finish reason, function calls) into the canonical ai_act schema.

Common deployment paths

  • Direct Gemini API for apps and agents
  • Vertex AI Gemini endpoints in EU regions
  • Multimodal inputs with commitment over media payloads
  • Function calling and code execution tool traces

Integration shape

@sigigo/providers-gemini — same core proof library as OpenAI, thin adapter for Gen AI SDK streaming and tools.

  • wrapGemini(client, sigigo) — drop-in client proxy
  • Shared session_id across multi-turn chats
  • Composable with LangChain GoogleGenAI chat models

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
  • ai_act.action_executed (function calling)
  • gpai_model_id (gemini model version)

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.