Summarise logs
Logs from the failing runs are summarised deterministically before any model sees them.
Agentic
CI failure triage
Read the failing logs, cluster them to shared causes, propose a fix, and let an engineer file it.
Teams report that eighty failing runs usually collapse to three underlying problems, and that finding them by hand takes hours. The pipeline summarises logs deterministically, clusters failures, and lets an agent propose causes and fixes that an engineer accepts.
The pipeline
Cluster
Group failures by shared signature; separate developer issues from platform issues.
Propose
An agent proposes a cause and a fix per cluster, with the evidence it used.
Engineer accepts
Nothing is filed or re-run until a person accepts the proposal.
What the record keeps
The failing runs, the log summaries, the clusters, the proposals, and what the engineer accepted.
Hours to minutes
Log reading and clustering happen before anyone opens a dashboard.
Owners, not noise
Failures are split into developer and platform issues and routed to the right team.
A person files it
Proposals wait for an engineer; nothing is re-run or filed on its own.
Connectors and tool coverage depend on the installed operators and your deployment.
See how failure can trigger a reviewed improvement cycle →Run it on your own infrastructure.
Apache 2.0 licensed, self-hosted. Talk about your CI and what a flaky day costs the team.