Skip to main content

​
What EvalGate is for

EvalGate is CI for AI behavior. It helps AI teams stop the same failure from shipping twice. You capture what your AI actually did, turn the bad cases into test cases, and then block any future change that brings those cases back.

​
The one idea behind everything

EvalGate is a single loop: trace → eval → gate.
  1. Trace. Your app records what the model did — inputs, outputs, tool calls, cost, latency.
  2. Eval. A person reviews a real failure and promotes it into a reusable test case.
  3. Gate. That case runs in CI. The next prompt, model, or retriever change that breaks it fails the build before it reaches users.
Everything else in EvalGate — dashboards, judges, playgrounds, monitors, reports — exists to make one of those three steps faster or more trustworthy.

​
What you do not need to know first

  • You do not need an account to start. The first gate runs locally.
  • You do not need to understand LLM judges, calibration, or online eval to read these guides. Each one explains only what it uses.
  • You do not need to be an ML engineer. If you can follow a numbered list, you can use EvalGate.

​
Up next

Learn how to find your way around the app.