Hiya Product Hunt 👋
I am Onyeka, and I have been obsessive about one downside: coding brokers can write nice code, however they may be able to’t shut a price ticket.
They omit CI screw ups. They do not know which factor to pick out up subsequent. They lose context on evaluate feedback. You find yourself babysitting them as an alternative of allowing them to send.
That is after spending a pair hours of your day atmosphere this all up.
**The issue** 👀
Each and every agent integration I noticed used to be the similar: sell off GitHub webhooks at your agent and hope for the most efficient. Uncooked log partitions. No construction. No routing. The agent burns context simply parsing noise.
What we constructed ⚡
AgentRail is an open-source process lifecycle API that provides coding brokers an actual dev loop:
– 📥 GitHub Problems and Linear tickets turn into routable duties mechanically
– 🔁 CI effects come again as structured summaries — no longer uncooked log partitions
– 🔍 PR evaluate feedback are ranked by means of severity (blockading vs. advisory)
– 🔑 In step with-agent scoped API keys — each and every agent best touches what it wishes
– 🏠 Runs totally in the neighborhood, no account required
**How to take a look at it** 🚀
npm set up @agentrail-core/cli — that is it.
Completey unfastened for self-hosted, contributions are greater than welcome!
Agentrail is within the very early phases, and in truth introduced previous than deliberate to make it for Vercel Day! So please, spoil it, roast the hell out of it, all comments is welcome/
Would really like to listen to from any person working Claude Code, Codex, or Cursor brokers at scale. What is the very first thing that breaks if in case you have a couple of process working? 🙏
