memex ecosystem
built on memex

docvault

agentic developer documentation knowledge base

Drop library docs, changelogs, and RFCs in.
Agents extract APIs, breaking changes, and patterns —
linking libraries and concepts into a searchable reference.

github.com/drader/docvault

The promise

Breaking changes flagged
before they break you.


$ cp react-19-changelog.md wiki/raw/docs/
 
# Tomorrow morning:
 
$ cat wiki/changelogs/react-v19.0.0.md
 
⚠ BREAKING useReducer signature changed
⚠ BREAKING ReactDOM.render() removed
✓ New: Server Components stable
✓ New: use() hook
 
# Breaking changes never buried in prose again.
The reality

Your stack evolves.
Your knowledge doesn't.

Problem 01

Breaking changes are buried

They live in a 3,000-word changelog. You read the highlights. The breaking part wasn't highlighted.

Problem 02

Library knowledge scatters

API notes in one doc, gotchas in a Stack Overflow tab, migration guide in a Notion page nobody updates.

Problem 03

Staleness is invisible

Your notes say v3. The library is on v5. You don't know that until something breaks in production.

Problem 04

Patterns stay per-library

Three libraries solved the same problem three different ways. You implement it a fourth time.

docvault solves all four.

Problem 01 — Breaking changes are buried

A changelog that buries
breaking changes is a liability.

The situation

Changelogs are written for humans skimming on release day. Breaking changes get one line between two feature announcements. By the time you hit the runtime error, you've forgotten you even read the changelog. Post-mortem always says: "it was in the changelog."

The fix

Every ingest explicitly extracts breaking changes into a ⚠ BREAKING block at the top of the changelog page — never in prose, always structured. The library's version history table is updated automatically. Run /query "what broke in react 19" and get a direct answer with citations.

Problem 02 + 03 — Scatter and staleness

One page per library.
Always current.

The situation

Every library in your stack has notes spread across browser bookmarks, Notion docs, team wikis, and your own memory. When a new version ships, nothing updates automatically. You discover staleness when the old API throws at runtime.

The fix

wiki/libraries/react.md is the single source of truth — current version, key APIs, gotchas, migration notes, and a full version history table. Every new changelog updates it automatically. When the queue is empty, the orchestrator scans for libraries with outdated entries and flags them in the journal.

Problem 04 — Patterns stay per-library

Patterns that span libraries
get a page of their own.

The situation

Optimistic UI appears in React docs, SWR docs, and TanStack Query docs. Each explains it from their own angle. You learn it three times and never build a mental model that transcends any single library.

The fix

Any design pattern demonstrated in a doc gets its own page in wiki/patterns/, linked from every library that uses it. Concept pages in wiki/concepts/ capture the cross-library idea. The Monday orchestrator writes synthesis pages when the same pattern appears across multiple libraries.

Wiki structure

One page per library. One page per release.

Input
  • raw/docs/
  • changelogs
  • API references
  • RFCs, tutorials
Agent Output
  • libraries/
  • changelogs/
  • patterns/
  • concepts/
Orchestrator
  • syntheses/
  • journal/
  • Staleness scan
  • Monday patterns

MCP hybrid search (BM25 + vector + RRF) over the full wiki.
Breaking changes always explicit. Library knowledge always consolidated.

Get started

Three commands.


$ git clone github.com/drader/docvault my-docvault
$ npm install && npm run setup
$ npm start
 
✓ Agents scheduled (Tue–Fri 09:00).
✓ Wiki seeded from your stack.
✓ Daemon running in background.
 
Drop any changelog, API page, or RFC into wiki/raw/docs/
— or run /ingest to process immediately.
 
github.com/drader/docvault
built on memex
TypeScript · Claude API · MCP
1 /