How it works

Your coding tools do the work. Canon defines how the work gets done.

Canon is a system for designing, validating, and compiling your software development process. You define the agents, roles, handoffs, conditions, engineering rules, and human approvals your team follows. Canon turns that model into the agents, workflows, instructions, and checks your existing tools run.

  1. 01 · Orchestrate

    Define

    Model roles, handoffs, conditions, and human gates on a visual canvas.

  2. 02 · Orchestrate

    Validate

    Find unreachable roles, dead ends, and loops. Simulate routing before it governs work.

  3. 03 · Orchestrate

    Compile

    Turn the model into the files your repository, CI, and coding agents execute.

  4. 04 · Command

    Run

    Start governed work in GitHub. Humans decide the gates the flow reserves for them.

  5. 05 · Command · Verify

    Supervise

    Observe handoffs, resolve exceptions, and check the repository against the process.

What Canon is

Canon builds the system that builds your software.

AI coding tools made individual developers far more capable and team delivery less consistent. Each developer ends up with different agents, prompts, review paths, standards, and approval rules. Canon gives the team one explicit, validated process for how people and agents work together.

A software factory, written down
The complete operating model that moves work from idea to release: product definition, architecture, development, review, testing, security, approval, deployment, and documentation. You include the stages your team actually uses.
Orchestration, not code generation
Canon does not write your application code. Coding agents in your flow do. Canon defines the system those agents, reviewers, testers, and human approvers work under, and coordinates how work moves between them.
A model, not a diagram
The canvas is how you design it. Underneath, Canon understands roles, handoffs, branching conditions, human stages, start and end states, escalation paths, and agent authority, which is what makes validation, simulation, and compilation possible.

From model to repository

Compile turns the process into files machines execute.

You design the process visually. Canon compiles it into structured instructions and configuration your repository, CI, and AI development tools can run. Every file comes from the same compile, so the diagram, the specification, and the executable workflow always describe the same system.

What the GitHub package contains
Agent definitions, workflow files, and prompts. Engineering instructions such as AGENTS.md. The machine-readable blueprint specification. Installation, verification, and pipeline documentation, with a diagram in PDF and SVG.
Where it runs
In your repository, on GitHub Actions. Canon does not ask you to move code into a Canon-hosted repository.
What happens after compile
Orchestrate installs the flow in a connected repository and keeps it current. Command starts governed work, routes gates to the right people, and records what happened.

What Canon is not

Canon sits above the tools you already use.

It does not compete with them, and it does not replace them.

Not another AI coding tool
Copilot, Codex, Claude Code, Cursor, Gemini CLI, Windsurf, Replit, Lovable, Bolt, and v0 perform development work. Canon defines how that work is performed, reviewed, handed off, approved, and completed.
Not a replacement for GitHub
GitHub stays your repository, collaboration environment, and CI/CD platform. Canon compiles the process into it.
Not GitHub Actions with extra steps
Actions is an execution mechanism. Canon models what sits above it: who or what performs each stage, what each stage receives and produces, where people intervene, and which rules pick the route. It then compiles that model into Actions workflows.
Not an AGENTS.md file
Instruction files are useful execution artifacts. On their own they carry no routing, conditions, human gates, validation, simulation, or versioning. Canon generates AGENTS.md as one output of the model.
Not a prompt
A prompt describes behavior. It is not a durable, versioned, inspectable process. Canon keeps the process separate from whichever model or coding tool executes it.

Portability

Your engineering process belongs to you.

Canon separates the process from the tool that executes it. Change models, coding agents, or AI vendors and the definition of how you ship stays the same. You recompile rather than redesign.

Portable files, in your repository
Every flow compiles to a JSON and YAML specification you own. The specification and the generated package live in your repository alongside the tools that run them, not only inside a Canon runtime.
GitHub is the full target today
GitHub gets the complete package of agents, workflows, prompts, instructions, and documentation. For other environments, the portable specification is the source of truth. Where a target has no equivalent for a Canon feature, the intent is kept in the specification and written instructions.
No lock-in to managed execution
A subscription pays for Canon to install, maintain, and operate the flow. Cancel, and the files Canon compiled into your repository stay there unchanged. Only Canon-managed orchestration pauses.

Who it is for

Any team where more than one person or agent ships code.

Small teams use Canon to get structure without management overhead. Larger teams use it to hold engineering rules and approvals constant across developers and projects.

Useful without AI, urgent with it
Canon models human and automated stages together. The more engineering work you delegate to agents, the more a consistent operating model matters.
You decide how you build
Templates are starting points, not a prescribed SDLC. Your organization decides the roles, agents, sequence, conditions, standards, human gates, and authority structure.
People keep authority
Human stages, approvals, escalation, security sign-off, product acceptance, and manual intervention are first-class parts of the model. An agent cannot proceed past a stage reserved for a person.

Design your process. Validate it before it governs anything.

Start from a template, describe how you build in plain English, or begin on a blank canvas. Design is free.