We built a maintainer agent, not a coding agent
The crowded part of the market is the coding agent — the thing that writes the diff. Copilot, Claude Code, OpenHands, Devin. They are good, they are getting better, and you probably already run one.
The un-crowded part is everything around the diff: triaging the issue, asking the reporter for a repro, deciding whether a fix is even in scope, watching the pull request through CI and review, and shipping the release. That is the work a maintainer actually does, and it is the work developerz.ai automates.
Thin orchestrator
developerz.ai is a thin orchestrator. It does not write your code. When an issue is triaged and qualified, it hands off to the coding agent you choose — on your keys — and then runs the loop around that agent: it opens nothing it cannot audit, merges nothing the machine gates have not cleared, and discloses on every comment that it is a bot.
That boundary is deliberate. Coding agents improve monthly; a tool that welds
itself to one of them ages badly. By staying the orchestration layer, we ride
whatever agent you already invested in — and swapping Copilot for Claude Code is
a one-line change to your .maintainer.yml, not a migration.
What the loop covers
Three loops, each bounded by your policy and audited end to end:
- Issue loop — read the issue, detect other bots, ask for a repro if your policy wants one, qualify or escalate, then hand off.
- PR loop — wait for CI green and review to settle, classify the risk of the change, and merge when the gates pass. Breaking changes and first-time contributors pause for a human.
- Release loop — cut the release when the merged PRs warrant it.
None of these is where the model writes code. All of them are where a maintainer spends their attention.
Why this is the honest design
Two of our non-negotiables fall straight out of the thin-orchestrator stance:
- BYOK only. The coding agent runs on your Anthropic, OpenAI, or Gemini key. We never proxy or resell tokens — your provider invoice is the whole bill.
- Audit is trust. Every decision, every tool call, every merge lands in an append-only log. If it isn't logged, it didn't happen.
We are not trying to be the best coder in your repo. You already have one. We are trying to be the maintainer who never sleeps, never skips the audit trail, and never merges past your gates.