Built
Sentinel
An autonomous engineer for production: detect, diagnose, reproduce, fix, verify and ship.
Sentinel watches live services through OpenTelemetry, spots anomalies, links each incident to deploys, commits and flags, and reproduces the failure in a sandbox. Agents then write a fix and a regression test, which is verified against real traffic and rolled out through a canary.
- Platforms
- Web, Docker
- Built with
- Python, OpenTelemetry, ClickHouse, tree-sitter, Claude API, Next.js
What it does
- Anomaly detection against a robust baseline
- Root causes ranked with evidence from commits, code, flags and past incidents
- Sandbox reproduction: the fix must fail before and pass after
- Canary rollout from 10% to 100% with automatic rollback, or a pull request with the incident report
How it’s built
- A demo breaks a sample shop and closes the whole loop in about 2.5 minutes
- About 95 unit and integration tests plus end-to-end tests