How it works
Your goal.
A factory on your host.
Hivemill is an open-source, self-hosted system that coordinates AI coding agents through implementation, review, QA and merge.
Follow the workflowWhat runs where?
Your laptop
Set the goal.
Connect to the host, delegate work and inspect progress.
Your Debian or Ubuntu host
The factory runs here.
Agent sessions, repository worktrees and the work record stay on the host.
Connected services
Models and GitHub.
Claude Code and Cursor connect to their model providers. GitHub holds pull requests and CI checks.
Your laptop can close while the host stays running and connected. Work may pause for decisions or resource and usage limits. Self-hosting the factory does not mean the models run locally.
The workflow
One goal.
A checked change.
Describe the outcome and acceptance criteria. The factory coordinates the handoffs and records what happened.
Follow a demo goal- PlanThe boss records the goal; a director and lead organize tasks and agents.
- BuildA coder works in a separate worktree and opens a pull request.
- CheckA reviewer checks the diff; QA tests the acceptance criteria.
- MergeBy default, both verdicts must pass before the judge merges. Your configured policy controls the final gate. Read the merge policy.
The work record
Inspect the evidence.
Follow a goal through its tasks, agent messages and verdicts. The read-only web dashboard keeps progress and decisions visible.
Browse demo goals
Make it yours
Adapt the workflow.
Reuse roles, skills and plugins from the catalog. Set resource limits and Claude usage caps for the host; use monitoring to inspect capacity and estimated costs.
Browse the catalog