Mission

We want to trace the frontier of AI, and use what we learn there to help shape where agent telemetry goes next.

AI moves fast. A new framework or model API comes out and people are running it in production a week later. If you’re the one responsible for that agent, you need to see what it’s doing now. Waiting a year for a spec to catch up isn’t an option.

Why OpenInference exists

OpenTelemetry is the right foundation for all of this. We build on it and we want it to succeed. The GenAI semantic conventions are being worked out in the open, across a lot of vendors, and that takes time. It should. That’s how you end up with a standard people can rely on.

The instrumentation can’t wait for that, though. So when a framework or provider starts to matter, we instrument it, ship it, and see what holds up once real people use it. Every instrumentor is our best answer to “what should a trace of this actually look like?”, and we find out quickly if we got it wrong.

How it feeds back into OpenTelemetry

OpenInference isn’t meant to be a separate thing off to the side.

OpenInference is where we figure things out quickly. OpenTelemetry is where the things that work end up.

Further reading