Get your first knowledge base running in under 10 minutes.

Prerequisites

Claude Code installed (v1.0+)
Node.js 22
A running Kanon server (self-hosted or cloud)
Your web application's source code in a local directory or GitHub repo

Step 1: Install the plugin

claude plugin install kanon

Or with a local build:

claude --plugin-dir /path/to/kanon/plugin/kanon

/kanon:setup

This opens a device-authorization flow:

The plugin gives you a verification URL and a code
Open the URL in your browser
Sign in or create an account
Enter the code to approve the device
The plugin stores your credentials locally at ~/.kanon/credentials.json

You'll also configure:

Server URL — your Kanon instance
Repo slug — e.g., your-org/your-app

The config is saved to .kanon/config.json in your project.

Step 3: Discover your product

/kanon:discover

Kanon reads your codebase and proposes a taxonomy — the domains and features that make up your product. It analyzes:

Route definitions (Next.js, Rails, Express)
Navigation components (sidebar menus, nav bars)
Middleware and auth guards
Data schemas (Prisma, Rails, SQL)
Module structure
i18n keys
Code ownership (CODEOWNERS)

The result is a set of proposed domains and features. Nothing is published — you review first.

Options

--refine — also crawl the running app in Chrome for visual navigation data
--auto-approve — skip human review (for CI or trusted pipelines)

Step 4: Review and approve

Open the review UI in your browser:

https://your-kanon-server/discovery/your-org/your-app

You'll see every proposed domain and feature with:

Name and description
Confidence score
Evidence (which code signals support the proposal)
Boundary globs (which files belong to this feature)

For each proposal: Accept, Edit & Accept, or Reject.

Accepting a feature materializes it — it becomes a real entry in the knowledge base that can be scanned.

Step 5: Scan a feature

/kanon:scan

The scan deep-reads the feature's code and generates:

Business rules and claims (with trust status: verified/asserted/drifted)
A comprehensive guide with narrative, principles, lifecycle, data model, routes, jobs, flags, events, security findings, test plans
Mechanical facts (tables, enums, routes, constants, events, experiments)

This is the expensive step — it makes LLM calls for extraction and synthesis. A typical feature takes 2-5 minutes and costs $1-5.

Step 6: Read the knowledge base

Open the knowledge base in your browser:

https://your-kanon-server/knowledge-base/{repo-id}

You'll find:

Overview — domains, features, coverage, honest gaps
Feature guides — deep-dive documentation for each scanned feature
Tracking plan — analytics event inventory
Flag plan — feature flag lifecycle
Experiments — A/B test inventory
Security audit — static analysis findings
Testing plan — E2E test scenarios
AI chat — ask questions about your product

Step 7 (optional): Keep a worker running

/kanon:worker

With a worker online, your team stops needing a terminal at all — the whole loop runs from the browser: Run discovery with worker on the review page (steps 3–4 above, no terminal), Scan all with worker once the taxonomy is approved (step 5), and Run with worker on any pending change for a review-ready PR. Every run streams a live session trace. See The worker daemon.

Next steps

Set up integrations — connect Linear, GitHub
Configure CI coverage — wire test results for trust verification
Use the AI chat — ask questions and draft tickets
Work on changes — implement pending changes from the KB
Run a worker — execute changes from the UI, no terminal

Quickstart