When CI goes red it says something broke, not why: OpenSRE traces the cause and pushes a fix for review
Synopsis
When continuous integration (CI) fails, it usually reports only that something failed, not why or how to fix it, and teams often re-run and move on. OpenSRE is an open-source CI/CD reliability agent that, when a run fails, reads logs, tests hypotheses and pushes a fix for review. It separates CI-caused failures from code-caused ones and can be self-hosted with your own models; the project is in public Alpha, with core workflows usable for early exploration but not yet fully stable.
Interpretation
OpenSRE hands post-failure investigation to a command-line agent: when a run fails it reads logs, tests hypotheses and pushes a fix for a person to review. Previously CI going red told a team only that something failed, not why or how to fix it; under deadline pressure teams re-ran, flaky checks piled up, main stayed red, and the real investigation waited for whichever senior engineer had time. The project holds itself to real-world end-to-end tests across cloud-backed scenarios and keeps semantic test-catalog naming so the boundaries between end-to-end and unit tests and between local and cloud stay obvious.
It separates failures caused by CI itself from those caused by code, so teams fix the right thing rather than just re-running. That separation rests on the agent reading the failed run's logs and testing hypotheses, not on the outcome of a re-run. The capability targets use from a terminal opened in a code project, with a Git repository recommended for the full workflow, and the installer covers macOS, Linux and Windows.
Perspective
The public material describes OpenSRE for users working from a terminal in a code project, recommends a Git repository to demonstrate the full workflow, and its installer covers macOS, Linux and Windows; the project describes itself as in public Alpha, with core workflows usable for early exploration but not yet fully stable, and APIs and integrations may still change.
The project describes itself as in public Alpha, with core workflows not yet fully stable and APIs and integrations possibly still evolving; the public material gives no quantitative results for fix success rate or investigation time.
