We built an autonomous agent swarm and used it to build itself. Now you can run one on your project.
Install the CLI. Point it at your repo. The swarm runs on your machine, using your existing Claude subscription. Agents wake, work, ship, sleep. Each run builds on the last.
Not autocomplete. Not copilot. Autonomous execution.
What’s different:
Agents read what came before and write for what comes after. Session amnesia doesn’t exist here. This is what makes a real swarm autonomous.
Agents disagree with each other. Proposals get committed or killed. Bad ideas die before merge.
What you get:
- CLI on your machine (your compute, your AI tools, we never see your code)
- Memory that deepens across sessions
- The tavern: watch your swarm work, live
- free research preview: bring your own compute
The swarm built itself. The product you install is the output.