Custos reads every issue, writes the fix, proves it in a sandbox, and merges when it’s safe. Here is what that looks like from your side of the screen.
Every new issue arrives with a verdict and a plan. Duplicates are closed with a pointer, in-scope bugs are queued for a fix, and anything ambiguous waits for you instead of an agent run.
Overnight, the queued bug became a branch, the branch became a diff, and the diff brought its own tests. The PR reads like a teammate wrote it: what changed, why, and how it was verified.
While it reads your code, Custos notices what a good maintainer would: the missing backoff, the query that fans out, the runtime aging out of support. Each becomes a proposal with its reasoning attached. Approve one and it goes straight to the solve queue.
Deliveries retry instantly and hammer failing endpoints. Seen in 40+ log incidents this month.
One request issues 41 queries. A single join drops p95 latency by an estimated 380ms.
Two breaking changes affect this repo. Migration plan drafted, 14 call sites to touch.
Curious how it got there? Open the log. Every command, test result, and decision is recorded from the orchestrator’s side of the glass, so what you read is what actually ran, not what the agent claims.
Merge policy lives in your dashboard, not in a prompt. Cap diff sizes, protect paths, require a human on public APIs. The orchestrator enforces every rule server-side, and the agent has no way around it.
The long tail of small, well-specified issues nobody gets to.
Bumps, breaking-change migrations, and the tests to prove them.
CVEs patched and verified the day they land, not next sprint.
Reproduces the flake in a sandbox, then fixes the race, not the symptom.
Finds untested paths and writes the tests your team skipped.
Keeps READMEs, changelogs, and API docs true to the code.
Connect a repository and the inbox starts sorting itself today.