Home / Platform / CI/CD for web & mobile

Test the product, not the endpoint

Calling the agent's API proves the model works. It does not prove the product does. Between the caller and the model sit microphone permissions, audio routing, echo cancellation, a websocket that drops on a train, and a mobile network that adds three hundred milliseconds each way. Every one of those has shipped a broken release for somebody. Prologue drives the shipped app the way a caller does, so those layers are under test too.

Web, iOS and Android, from one suite

The same tests run against all three surfaces. A regression that only appears on Android, or only on a cold start, or only when the app is backgrounded mid-call, shows up as a failing check next to the commit that caused it rather than as a support ticket a week later.

Wired into the pipeline you already have

Runs trigger on every deploy and report back as checks, next to prompt and code review. A prompt change is a deploy like any other, and it gets the same gate: the suites run, the scores come back, and a drop blocks the merge instead of reaching production first.

What a run tells you

• Which surfaces passed, and which failed on which turn.
• How the scores moved against the previous run, per dimension.
• The conversation itself, so the failure can be read rather than guessed at.

Built on your own failure modes

The suites that run here come from evals on real conversations and from the personas and custom tests written for the situations your agent actually meets.