Hermes fixes failed CI runs
A workflow fails at 2am, and by morning there’s a tested pull request waiting for your review.
Hermes
is Nous Research’s open-source agent: a terminal agent plus
hermes gateway, a service that listens on messaging platforms and
webhooks. Its docs already cover posting GitHub Actions
failures to chat. This plan pushes one step further than notification. A
failed workflow fires a webhook, a Hermes route turns it into a prompt, and
the agent pulls the logs with gh, fixes the code in a
Docker sandbox, runs the tests and opens a PR. You review the
diff. Nothing here ever merges.
Effort and cost. An evening if you already have a VPS or homelab box running Docker. A weekend from a laptop behind NAT, because GitHub has to reach your gateway. Every failure is an LLM run over CI logs, so point it at a cheap model. Hermes takes any model with your own keys. The traps page does the arithmetic.
The plan in six parts
Is this for you?
Yes, if you have a solid test suite and most failures are
code failures: an import broken by a refactor, a fixture that
drifted, a nightly build someone’s Friday commit killed. Even the
diagnose-only version (no PR) is worth having. Skip it if
most failures are infra flakes, where a retry button is cheaper, or if you
can’t host an endpoint GitHub can reach. It covers one GitHub repo.
GitLab (the gh glue only fits GitHub), monorepo routing and
auto-merge are out. For picking a terminal agent, see
№ 1 in this series. I checked all of
this against the Hermes docs at v0.19.0, July 2026.