The governance runtime for AI-assisted engineering. Keep AI agents bounded, inspectable, and safely restricted to approved work zones.
- 🚫 No Opaque Loops: You control exactly when agents plan, run, and publish.
- 🛡️ Bounded Execution: Agents operate strictly within approved risk and zone limits.
- 🔍 Inspectable State: Every decision, approval, and output is captured as durable evidence.
- 📖 Governed Packets: Turn unstructured chat into canonical, versioned markdown artifacts.
Canon operates on a simple, predictable four-step mental model:
init-> Prepare the workspace.run-> Start a governed session with explicit boundaries.approve-> Review and unblock the agent when human judgment is needed.publish-> Promote the final artifacts into your repository's permanent memory.
Get your first governed session running in seconds:
curl -fsSL https://apply-the.github.io/packages/apt/gpg.key \
| sudo gpg --dearmor -o /usr/share/keyrings/apply-the-archive-keyring.gpg
echo "deb [signed-by=/usr/share/keyrings/apply-the-archive-keyring.gpg] https://apply-the.github.io/packages/apt stable main" \
| sudo tee /etc/apt/sources.list.d/apply-the.list
sudo apt update
sudo apt install canonbrew tap apply-the/canon && brew install canonDownload the installer directly from the Releases page. Alternatively, use Windows Subsystem for Linux (WSL) and follow the Linux instructions.
- GitHub Release (.deb fallback): Download
.debfiles directly from the Releases page. - Source install fallback (requires Rust): Run
cargo install --path .from the repository root.
Verify canon was installed correctly:
canon --versioncd my-project
canon init
canon run --mode requirements --risk bounded-impactIn supported interactive terminals, canon init now opens a guided assistant
selector by default. Use canon init --non-interactive for scripts, CI, or
machine-readable output such as --output json. The guided selector includes
Codex, Copilot, Claude, Cursor, and Antigravity.
The public documentation is aligned with 0.69.0. Where the site links back
to repository source, it now points at the 0.69.0 release line.
Canon now publishes governed_reasoning_posture_v2 as the current stable
reasoning-posture contract for downstream consumers. The new line keeps Canon
as the semantic owner while making selector shape, independence minima,
confidence handoff, provenance, compatibility windows, and active-versus-legacy
migration rules explicit and fail-closed.
Backlog is also a delivered governed mode in this line. A successful backlog
run emits the full planning packet and may add execution-handoff.md when one
slice is credibly ready for downstream implementation. When planning is
complete but slice evidence is still weak, Canon keeps the full packet and says
handoff unavailable instead of inventing execution readiness.
These are the commands you'll actually use every day:
| Command | What it does |
|---|---|
canon run |
Start a new governed session with explicit boundaries. Available modes include requirements, architecture, backlog, brainstorming, debugging, change, incident, and more. |
canon status |
See exactly what the agent is doing right now. |
canon inspect |
Review generated evidence and artifacts. |
canon approve |
Unblock a session that hit a governance gate. |
canon publish |
Commit the final work into your repository. |
For advanced integrations, semantics, and architecture, explore the tech-docs/ folder:
Want to build or develop Canon itself? See CONTRIBUTING.md. Use the GitHub issue templates under .github/ISSUE_TEMPLATE/ when reporting bugs or feature requests. For vulnerabilities, follow SECURITY.md.