Skip to main content
1

Initialize fracta in your project

From inside any git repository:
This creates a .fracta/ directory for state and worktrees. It does not modify your repo’s tracked files.
2

Spawn an agent

Fracta creates a worktree under .fracta/worktrees/hello, launches your default runtime CLI in that worktree, and tracks the agent in its state store.Useful flags:
3

Watch progress

Streams the agent’s events (tool calls, output, status transitions) to your terminal in real time. Ctrl-C to detach without killing the agent.
4

See the result

Prints the agent’s recent semantic output. Run after the agent has completed (status completed in fracta list).
5

Clean up

fracta kill removes the agent’s worktree and state entry. The feature branch in the underlying git repo is also deleted. Make sure you’ve merged anything you want to keep.

Spawn multiple in parallel

Then list them:
Each agent runs independently, in its own worktree. They can message each other via fracta_send if you ask them to.

What’s next

Pick a Deployment Mode

Local-process vs docker-compose vs kubernetes.

Configure Your Runtime

Claude, Codex, and OpenCode setup.

CLI Reference

Every command and every flag.