Beta — early access is invite-only right now. Request access →
Use case · AI software team

Stories in, pull requests out.

A team of agents on GitHub: stories become tasks for each role, a coding agent opens the pull request, and a reviewer agent comments. A person approves the merge.

Pull request opened by the coding agent
A pull request opened by the coding agent, with its own summary.
How it runs

From a story to a reviewed pull request.

01 Plan

Tasks per role

A crew turns a story into tasks for each role, using a prompt the owner can edit.

02 Build

A coding agent works

Claude Code runs on a GitHub runner with the model from the workspace's own AI providers.

03 Review

A reviewer comments

A reviewer agent reads the diff and comments; a person approves and merges.

Task record

Every task is a record.

Tasks live in an entity, so status, the pull request link, the agent used and the prompt are queryable like any other data.

Task record
A task record with its pull request and status.
Bring your own agent

Swap the coding agent with one file.

Claude Code, Codex, Copilot, Jules and Muse Code are connectors that follow the same start-task and get-task contract. Adding another is one connector file.

Reviewer agent comment on the pull request
The reviewer agent's comment. Approval and merge stay with a person.
Ready when you are

Describe it once. Run it forever.

Connect your AI client, build your backend, ship to production.