My workIn progress
AgentReplay
Record every step of an AI run, replay it on a timeline, fork any step with a changed prompt or model, and save a failed run as a test.
Companion to AgentGuard: replay what went wrong, fix it, then undo the original actions.
The problem
When an AI run goes wrong, you need to see what happened, try a fix, and keep the failed run as a test.
How it works
- 01
Record every step of an AI run.
- 02
Replay the run on a timeline.
- 03
Fork any step with a changed prompt or model.
- 04
Save a failed run as a test.
Progress
- Sign-up
- SDK and ingest
- Runs list and replay screens
- Fork / fix / undo — in review
- Tests and suites — next
Highlights
- Published SDK on npm: @sritarak/agentreplay (no runtime dependencies)
Built with
ReactViteTailwindNode.js 22ExpressMongoDB