Skip to content

Quickstart

Get Docuflow running in your project in 5 minutes.

Step 1 — Install

npm install -g @doquflow/cli

Step 2 — Initialise your project

Run this from your project root:

cd /your/project
docuflow init

This creates:

  • .docuflow/ — Wiki directory (sources, wiki pages, index, log, schema)
  • CLAUDE.md — Auto-generated agent discovery file (tells Claude, Copilot, Cursor which tools are available)
  • .docuflow/schema.md — Domain configuration template

After init, you'll see domain-aware next steps from docuflow suggest automatically.

Interactive setup

Run docuflow init --interactive for a guided setup that configures your domain (Code/Research/Business/Personal) with tailored schema and prompts.

Step 3 — Add your first source

Drop a document into .docuflow/sources/:

cp /path/to/your/design-doc.md .docuflow/sources/

# Or write one directly
cat > .docuflow/sources/overview.md << 'EOF'
# My Project Overview

This is a TypeScript REST API with PostgreSQL...
EOF

Step 4 — Ingest into the wiki

docuflow ingest overview.md

Docuflow reads the source, extracts entities and concepts, creates wiki pages with context paragraphs, and updates the index.

To ingest all sources at once:

docuflow ingest --all

Step 5 — Ask questions

docuflow query "How does authentication work?"

Returns a synthesised answer with citations from your wiki pages. The more sources you've ingested, the better the answer.


What's next

Your AI agent can now discover Docuflow automatically via CLAUDE.md. In Claude, Copilot, or Cursor, you can start asking:

"Use Docuflow to search the wiki for authentication patterns."

"Ingest the new design document and tell me what changed."

Check wiki health anytime:

docuflow status     # quick overview
docuflow doctor     # full diagnostic

Open the web interface:

docuflow ui         # opens http://localhost:48821

CLI Reference → MCP Tools →