docforge

structured documentation infrastructure for humans and agents.

docforge crawls, renders, versions, and caches live documentation — turning messy, hard-to-scrape docs into clean, searchable artifacts that both humans and agents can reason over.

agents get a stable http interface.
humans get a readable, inspectable UI.
docs stop being scraped repeatedly and start being infrastructure.

---

why this exists

live documentation is one of the worst inputs for agents:

• js-heavy pages
• inconsistent structure
• high token cost to scrape repeatedly
• no versioning or freshness guarantees

docforge fixes this by:

• rendering docs once (properly)
• extracting structure, not just text
• storing versioned docsets with diffs
• exposing a deterministic api agents can trust

---

what docforge is

• an agent-native api for documentation
• a docset store with versioning + freshness
• a human-readable ui to inspect what agents actually see
• infra, not a chatbot

---

architecture (high level)

• next.js
  human ui + api gateway (agent entrypoint)

• fastapi workers + playwright
  render + crawl js-heavy documentation

• redis + bullmq
  async ingestion and crawling jobs

• postgres
  metadata, versions, chunks, search

• s3-compatible storage (r2 / minio)
  raw snapshots + extracted artifacts