Agentic developer documentation knowledge base — built on memex
Drop library docs, changelogs, and RFCs in. Agents extract APIs, breaking changes, and patterns — linking libraries and concepts into a searchable reference that grows with your stack.
- Drop a doc — Place a changelog, API reference, RFC, or tutorial into
wiki/raw/docs/. The agent identifies its type automatically. - Agent routes and extracts — The researcher agent reads the doc, identifies the library, version, and content type, then extracts APIs, breaking changes, patterns, and gotchas.
- Breaking changes are never buried — Every breaking change gets its own explicit flag in the changelog page — never hidden in prose. The library page's version history table is updated automatically.
- Orchestrator tracks staleness — When the queue is empty, the agent scans for libraries with outdated version entries and flags them in the journal.
wiki/
raw/docs/ drop changelogs, API refs, RFCs, tutorials here
libraries/ one page per library or framework
changelogs/ one page per version (library-vX.Y.Z.md)
patterns/ design patterns and architectural approaches
concepts/ programming concepts (memoization, SSR, etc.)
syntheses/ cross-library comparisons, migration guides
archive/ retired pages (never deleted)
- Node.js ≥ 20
- Claude CLI (
npm install -g @anthropic-ai/claude-code) obsidian-hybrid-search(npm install -g obsidian-hybrid-search)
git clone https://github.com/drader/docvault my-docvault
cd my-docvault
npm installnpm run setupRuns a 7-question intake interview. You list your stack, doc types you track, and the "I always have to look this up" questions in your codebase. The agent seeds stub pages for every library you name and Bootstrap Hypotheses based on your recurring questions.
npm start # foreground
npm run dev # watch mode
npm run install-daemon # background daemon (launchd on macOS, systemd on Linux)/ingest
Or drop any file into wiki/raw/docs/ — changelogs, API pages, tutorials, RFC text files. The daemon picks it up on the next cycle (Tue–Fri 09:00 by default) and routes it to the right wiki folder automatically based on content type.
| Command | What it does |
|---|---|
/ingest |
Process the next uningest doc in wiki/raw/docs/ |
/query |
Ask a technical question, get a wiki-backed answer with source citations |
/lint |
Check wiki health: outdated versions, missing changelogs, orphan pages |
/setup |
First-time intake interview |
/session-start |
Load context, log session open |
/session-end |
Flush scratch notes, log session close |
MIT