Hivemill

When you hand Hivemill a goal, AI coding agents build it, separate agents review and test it, and a judge decides the merge, all on a Linux server you control. You describe the outcome to a boss agent, a director staffs a lead, and the lead splits it into tasks, each with its own git worktree and branch. A coder opens a pull request and stops, because it cannot review, test, or merge its own work. The reviewer reads the diff, QA runs real tests against your acceptance criteria, and by default the judge merges only when both verdicts pass; anything else returns to the coder with findings. Pull requests, CI checks, and merges run through the GitHub CLI. Claude Code is the default runtime, a Cursor Agent org can run beside it, and hooks enforce each role's prompt and permissions. A spawn gate queues new agents until memory, load, and your Claude usage windows allow, so one modest machine stays responsive. Scheduled jobs mail the director recurring chores with a token cap per run, and a daily retro agent studies rejected work. Typical first jobs are bounded changes such as a filtered CSV export, a mobile navigation fix, or a duplicate-order check before import. A read-only web dashboard tracks goals, tasks, live agents, throughput, quality, and estimated cost per task. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.

Hivemill
Hivemill
Hivemill
Hivemill
Hivemill
Hivemill
Hivemill
Hivemill
Hivemill
Hivemill

Benefits

  • Close your laptop, keep shipping
  • The factory runs on your own Debian or Ubuntu host, and optional systemd boot units resume the boss, director, and active team leads after a reboot, so work continues while you are away.
  • Independent review before every merge
  • The coder opens a pull request and stops. A separate reviewer reads the diff, QA runs tests against acceptance criteria, and the judge merges only after both verdicts pass.
  • Spend and capacity stay under control
  • A spawn gate queues agents until RAM, load, and Claude five-hour and weekly usage allow, per-goal token limits stop new work, and Claude tasks show estimated cost at list prices.
  • A full audit trail of every change
  • Goals, plans, tasks, messages, verdicts, and decisions are stored with an append-only event log and exported to readable markdown every minute, so you can trace how any change was made.

Features

  • Fixed roles enforced by hooks
  • Boss, director, lead, PM, coder, QA, reviewer, judge, and cleaner roles each have their own prompt and permissions, enforced by Claude Code hooks and a worker guard.
  • Claude and Cursor orgs
  • Run a Claude Code org and a Cursor Agent org side by side, each with its own boss, director, leads, and workers, sharing one agent cap and resource gate.
  • Read-only web dashboard
  • Start factory serve for Overview, Goals, Tasks, Agents, Throughput, Quality, Tokens, Changelog, and Resources pages with org filters, light and dark themes, and JSON endpoints.
  • Machine-checked merge gate
  • QA verdicts list commands that must exit 0 on the PR head; the judge refreshes stale branches, waits for green CI, and merges with a head-commit match.
  • Scheduled jobs and retros
  • factory schedule sets up to five recurring jobs with a token cap per run, while a daily retro agent reviews rejected work and a scout vets new skills.
  • Opt-in worker sandbox
  • Linux user and mount namespaces can hide the owner's home and logins from reviewer, QA, and coder workers, using separate revocable tokens; it is a tripwire, not a boundary.

Apps Similar to Hivemill