Kanon reads your codebase and writes down what your product does, with the line of code behind every sentence.
Read it. Ask it. Change it.
Free to start. An engineer connects it in three commands. It reads the code in their own Claude Code session, and nothing leaves the machine but what you approve.
Your code is the only honest description of your product. Nobody who needs it can read it.
Every wiki is a snapshot of what someone believed six months ago. The one description that stays current is the code, written in a language most of your team can’t read.
The answer used to cost a Slack thread, two interruptions, and a day. Now it costs a question.
Kanon connects to your repository and derives a guide for every feature. Each sentence is a claim, anchored to the exact line behind it. When the code changes, the guide is read again.
A refund can never exceed the amount originally captured.
A failed charge is retried up to three times before the subscription is marked past due.
Checkout is gated by two feature flags; the one-click path is hidden when express_checkout is off.
We never write your tests. We find the ones you already have, and check what they actually prove. That’s the difference between “the tests pass” and “the rule is proven.”
A real test in your own suite exercised this rule, with the evidence and a date attached. Nothing else earns the word.
Inferred from the code and labelled honestly, the calm and neutral default. Most of a young knowledge base lives here, and that’s the truth, not a gap.
A fully-amber bar is the truthful picture, not a broken one. The green is believable precisely because we didn’t paint everything green.
Every flag, what it gates, whether it's on or off, and the exact line that declares it. The flag that's been on for eighteen months that nobody dares delete is in here, with its blast radius.
| Flag | Gates | State | Source |
|---|---|---|---|
| express_checkout | One-click checkout path | off | flags.ts:12 |
| new_pricing | Pricing table & tax lines | on | flags.ts:19 |
| legacy_dunning | Old retry cadence (no reads in 18mo) | on | flags.ts:27 |
| audit_log | Write-ahead audit trail | on | flags.ts:31 |
Product teams have always written requirements that engineers translate. This one doesn't need translating. A request written against the guide becomes a ticket, or a pull request, linked to the exact claims it changes. You review it before it's real.
“What’s the refund limit?” Answered without reading Ruby or waiting on an engineer.
“Why did this charge fail?” The exact retry and dunning behavior, cited.
“Show me the rule and the line that enforces it,” with its trust state.
Onboarding used to take a quarter. Now it’s an afternoon of asking questions.
Documentation decays every sprint. This compounds every sprint: the first source of truth whose maintenance cost goes down over time.
Kanon reads your repository from your team's own Claude Code session, so your subscription runs the crawl and nothing leaves the machine but the taxonomy you approve. Sign-in is a browser approval, no credentials touch the agent, and you review every proposal before a single feature is scanned deeply.