About
Your product’s behavior,
written down from the code
Kanon reads your source code and produces a verified knowledge base of what your product actually does. Every sentence cites the code behind it. Every pull request keeps it current.
The problem
Every software company has the same hidden cost: the knowledge that lives only in engineers’ heads. When a founder asks “what happens when a payment fails?”, the answer takes a Slack thread, a meeting, and a code dive. When a PM writes a spec, it’s outdated before it ships. When a new engineer joins, they spend weeks asking questions that the code already answers.
Documentation doesn’t solve this. People write it once and never update it. Within weeks, the docs say one thing and the code does another. The documentation becomes a liability — confidently wrong is worse than absent.
What Kanon does differently
Kanon doesn’t ask anyone to write documentation. It reads the code and writes it down. Every claim about your product — “the daily repayment cron runs at 20:00 UTC”, “a returned payment incurs a $40 fee” — is extracted from the source, with a citation linking back to the exact line.
When someone changes the code, the knowledge base detects it. A claim that was verified by a passing test and is now contradicted by a code change shows as drifted — visibly, honestly, without hiding the discrepancy.
The result is a knowledge base that a CEO can read to understand what the product does, a PM can query to plan features, an engineer can trust when debugging, and a QA team can use to build test plans. All from the same source: the code.
Principles
- Verified, not documented
- "Verified" is a formal state: a test in your own suite proved the rule. It is not a mood word. Kanon never calls something verified unless a passing test exercises the exact behavior.
- Visible honesty over silent confidence
- When the knowledge base doesn't know something, it says so. Uncited paragraphs are rendered at lower opacity. Drifted claims show the drift reason. An empty section is better than a fabricated one.
- Proposals, never assertions
- Discovery proposes a taxonomy for human review. The AI chat drafts tickets for human approval. Nothing is published or created without a human deciding it's right. The trust gate is the product's value.
- Code is the source of truth
- Every fact in the knowledge base traces back to a file and line number. If the code changes and the knowledge base doesn't, the discrepancy is surfaced, not hidden.
See what your product actually does
Point Kanon at a repository and read the knowledge base it generates.