wikipilot scans your codebase and generates a wiki where every page tracks its own source files — so when the code changes, the docs tell you instead of quietly going stale.
Docs get written once. The code they describe keeps changing forever.
Nothing tells you which pages are still true and which quietly rotted months ago.
By the time someone notices, they've already shipped the wrong instructions.
wikipilot reads your repo's structure and source content — no config, no annotations required to get started.
Every page is written with a record of exactly which source files it came from and when.
When those source files change, the page flags itself stale instead of quietly going out of date.
Every page tracks the exact source files it was generated from and when. Change the code and the page marks itself stale instead of quietly lying to the next reader.
A chat widget grounded only in your wiki's own content — no hallucinated answers from the open web, just what's actually documented.
Client-side search across every page, no server or index to run. Type and go, even on a static host.
Architecture and flow diagrams render straight from fenced code blocks in your Markdown — no separate diagramming tool.
wikipilot init scaffolds a Claude Code skill into your repo, so keeping the wiki in sync is one prompt away instead of a manual chore.
Build once, get plain HTML/CSS/JS. Serve it from anywhere — no runtime, no database, no vendor lock-in.
sources — the exact files this page was generated from.
last_synced — the timestamp it was last regenerated against those files.
stale — flips automatically once a source file changes underneath it.
The CLI is MIT licensed and stays that way. Only the hosted tier is a paid product.
Point wikipilot at a repo and see the freshness contract for yourself in under five minutes.