Rig is the cryptographically governed control plane for local AI coding.
Most AI coding tools mutate your code blindly. Rig forces AI work through isolated Git worktrees, cryptographic receipts, validation gates, and explicit review before anything touches your main branch.
Models propose. Rig disposes.
python3.14 -m pip install -e ".[dev]"rig init
rig ui
rig run --task fix-imports --provider custom-command- CLI: Use the terminal for scriptable, deterministic workflows and repository management.
- Windowed UI (
rig ui): Use the rich, windowed control plane for interactive work, streaming logs, agent chat, and governance gates. - Textual TUI: Retired. Use
rig uifor a rich interface or CLI commands for terminal workflows.
- Workspaces as the project authority boundary
- Agent lanes as governed children of a workspace
- Isolated worktrees
- Receipts
- Proposals
- Review and apply gates
- Job queue
- No silent mutation of main
- No auto-apply
- No provider direct mutation
- No background daemon by default
- All orchestration leaves receipts and logs
rig initrig config inspectrig runtime listrig model listrig provider listrig context buildrig system inspectrig job createrig job runrig run --task <task-id> --provider <provider_id>rig agent proposerig workspace statusrig workspace lanesrig workspace projectionrig workspace receiptsrig workspace recommendrig workspace createrig workspace reviewrig workspace applyrig log listrig log showrig debug bundlerig ui
Rig is usable as a standalone CLI and governance shell. Runtime/model/provider integration is advisory only and remains behind explicit policy gates.
Rig requires Python 3.14 or newer.
See docs/index.md.
Workspace and lane architecture starts in:
Rig originated from a migration out of the Anigma workspace, but the product surface is now standalone. Migration history remains in docs/migration/ for reference only.