Skip to main content

Observra

Observra is an AI Ops and Observability platform. It sits as a lightweight gateway between your application and the AI providers you already use (OpenAI, Anthropic, Gemini, Azure OpenAI, Ollama, and others), and turns every request into an immutable observation you can search, chart, and alert on - without asking you to hand over your provider credentials or change your client code.

Who it's for​

Organizations running AI-powered applications in production - engineering teams, platform/ops teams, and the leadership overseeing AI spend and risk - who need one governed source of truth across multiple providers, teams, and environments: How many requests are we making, and by whom? What's our p95 latency? What's this costing us, broken down by project and business unit? Did our error rate spike after that deploy? Are we within budget and policy? Observra is built for that - not as a replacement for your provider's own console, but as a single, enterprise-ready place to see usage, cost, latency, reliability, and compliance across every provider and every application in the organization.

How it works, in one sentence​

You point your existing provider SDK's baseURL at the Observra Gateway and keep sending your provider API key exactly as you always have; the gateway forwards your request untouched, measures it, and turns it into an observation in your dashboard.

Core ideas​

  • You keep your own provider SDK. No new SDK to adopt, no vendored client library. Configure baseURL and a Gateway Key, keep using openai, @anthropic-ai/sdk, or whichever client you already have.
  • Your provider credentials are never stored. Provider API keys are forwarded upstream on every request and never persisted or logged. See Gateway Authentication.
  • You control retention. By default, Observra stores only metadata (tokens, cost, latency, status) - not prompt or response bodies. Capturing full payloads is opt-in per Environment.
  • Your AI traffic always comes first. Observability is asynchronous: if the analytics pipeline degrades, your requests to the provider are unaffected.
  • One multi-tenant hierarchy. Organization → Project → Environment → Application. Gateway Keys are issued per Application per Environment.

Where to go next​

  • New to Observra? Start with the Quickstart - get a Gateway Key and see your first observation appear in under ten minutes.
  • Integrating the OpenAI SDK specifically? See OpenAI SDK Integration.