OpenRouter

OpenRouter Audit Logging & Multi-Model Evidence

Multi-vendor routing with one evidence boundary

OpenRouter fronts dozens of model providers behind an OpenAI-compatible API. Sigigo wraps the client so every routed call records the actual model used, fallbacks, and commitments — regardless of which vendor answered.

Where Sigigo sits

At the OpenRouter client boundary — the same OpenAI-compatible surface most apps already use. Sigigo records which upstream model served each request, routing decisions, and tool calls without requiring per-vendor wrappers.

Why teams route through OpenRouter

Production stacks increasingly abstract the model vendor behind a gateway:

  • One SDK surface across OpenAI, Anthropic, Google, Mistral, and others
  • Automatic fallbacks when a provider is down or rate-limited
  • Cost and latency routing without rewriting application code
  • When Article 12 applies, recording which model served each decision remains part of a reconstructable log — Sigigo maps that into gpai_model_id

GPAI traceability

Sigigo maps OpenRouter response metadata — model id, provider, and version — into gpai_model_id on every ai_act event. Auditors see the actual inference chain even when your app only knows the OpenRouter model slug.

Integration shape

Drop-in wrapper over the OpenRouter SDK or any OpenAI-compatible client pointed at OpenRouter.

  • wrapOpenRouter(client, sigigo) — same pattern as direct OpenAI
  • Composable with Vercel AI SDK via @openrouter/ai-sdk-provider
  • EU preset: commitment mode + public anchor off
  • No duplicate events when stacked with LangChain on the same session

Evidence events

Canonical event types emitted through this integration path:

  • ai_act.inference_started / inference_completed
  • ai_act.input_received (Private Commitment Mode)
  • ai_act.output_generated
  • gpai_model_id from routed provider response
  • fallback / routing metadata on provider switch

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.