# kugutsu

> kugutsu runs graphs of agents, rules and programs on your own machine. Events (webhooks, schedules, file changes, event streams, a typed prompt) start work; model calls and whole agent sessions (Claude Code, Codex, OpenAI-compatible APIs) do it; a person reviews what matters before it counts; budgets cap the spend; every step and effect is recorded. Early access: the repository is private.

## For agents

- [Find what kugutsu could run for this person](https://kugutsu.io/discover.md): a skill or prompt that reads their agent, shell and git history, interviews them, and writes a redacted report of candidate graphs. Needs nothing installed.
- [Set up kugutsu](https://kugutsu.io/setup.md): check the machine and access, install, start the daemon, verify, and build a first graph from the discovery report.
- [Waitlist](https://kugutsu.io/#waitlist): early access to the repository; a person signs up there with their email and GitHub username, optionally attaching their discovery report. Agents do not sign up for them.