DeepDoc
v2.3 — inline validation · AI chatbot

Engineering docs
your team will
actually read.

DeepDoc generates internal documentation from your codebase — architecture, endpoints, integrations, runtime surfaces — and ships a chatbot that answers from real source with exact file paths and line ranges.

View on GitHub

Python 3.10 · Node 18+ · MIT license · Any LLM provider

bash
Works with Python TypeScript JavaScript Go PHP Vue OpenAPI
See it in action

Watch the pipeline run.

43 sec · all five phases
What it does

One command. A docs site
your team won't ignore.

01

Bucket-based generation

Architecture, endpoints, integrations, and schemas planned as cohesive docs — not one noisy page per file. The planner classifies your repo, proposes buckets, assigns files and symbols to a final reader-first structure.

02

Evidence-grounded chatbot

Answers hydrated from indexed source with exact file paths and line ranges. Generated docs serve as references, not proof. Three modes: fast lookup, deep research, and code-first.

03

Incremental updates

`deepdoc update` diffs against the last synced commit and regenerates only affected buckets. Smart enough to do a targeted replan when new integrations or endpoints appear.

Pipeline

Five phases. No magic.

Every page is grounded in scanned evidence and MDX-compiled before it hits disk. If compile fails, DeepDoc reprompts with the exact error.

  1. 01
    Scan
    Parse source. Detect endpoints, runtime surfaces, integrations, OpenAPI specs.
  2. 02
    Plan
    Multi-step LLM planner. Classify repo, propose buckets, assign files.
  3. 03
    Generate
    Build evidence packs. LLM call per bucket. MDX compile-check before write.
  4. 04
    API Ref
    Stage OpenAPI assets into Fumadocs /api/* pages when a spec exists.
  5. 05
    Build
    Write site scaffold, page tree, nav, search, and optional chatbot widget.
Getting started

Three commands.

Initialize against any Python, JavaScript/TypeScript, Go, PHP, or Vue codebase. Set one API key. Generate. The chatbot inherits the same key automatically and runs embeddings locally via fastembed.

Read the full docs
bash
# 1 — install $ pip install deepdoc # 2 — initialize against your repo $ deepdoc init --with-chatbot --provider anthropic $ export ANTHROPIC_API_KEY=sk-ant-... # 3 — generate and preview $ deepdoc generate ✓ 47 pages generated in 1m 12s ✓ Chatbot index built (1,247 chunks) $ deepdoc serve Docs → http://localhost:3000 Chat → http://localhost:3000/ask

Your codebase already tells a story. Let it.

MIT licensed. No account. No cloud. Runs entirely in your terminal.