Wiki-Forge

Auto-generate knowledge base from code & documents.

Wiki-Forge indexes your codebase and documents into a structured, interlinked wiki — then lets you query it via Web UI, CLI, or AI agents (Claude Desktop, Cursor).

Features

Feature	Description
Code-to-Doc	Auto-generate wiki from source code (19 languages)
Document Ingestion	PDF, DOCX, PPTX, HTML, Markdown → wiki pages
Q&A Chat	Ask questions in natural language, get answers with citations
Knowledge Graph	Interactive graph visualization (Sigma.js)
Drift Detection	Auto-detect code ↔ wiki contradictions (7 types)
MCP Server	Claude Desktop / Cursor integration (5 tools)
Multi-Project	Unified wiki across multiple repos
CGC Engine	Rust-accelerated code intelligence (Tree-sitter + KuzuDB)

---

Quick Start

Install

# Full install (recommended)
pip install "wiki-forge[all]"

# Or minimal (CLI only, no server/MCP)
pip install wiki-forge

1. Index a codebase

cd my-project

# Free index (CGC only, no LLM cost)
wiki-forge init --cgc --no-llm

# Full wiki generation (uses LLM ~$0.04)
export GEMINI_API_KEY=your-key
wiki-forge init --cgc

2. Query

wiki-forge query "how does authentication work?"

3. Web UI

wiki-forge serve --port 5757
# Open http://localhost:5757/wiki/

4. AI Agent (Claude Desktop / Cursor)

{
  "mcpServers": {
    "wiki-forge": {
      "url": "http://localhost:5757/mcp/sse"
    }
  }
}

---

Install Options

pip install "wiki-forge[all]"          # Everything
pip install "wiki-forge[server]"       # Web UI + REST API
pip install "wiki-forge[cgc]"          # CGC Rust code intelligence
pip install "wiki-forge[mcp]"          # MCP server (Claude/Cursor)
pip install "wiki-forge[converters]"   # PDF, DOCX, PPTX converters

---

CLI Commands

Command	Description	LLM?
wiki-forge init [--cgc]	Index repo + generate wiki	Yes
wiki-forge init --cgc --no-llm	CGC index only (free)	No
wiki-forge update [--cgc]	Incremental update (changed files only)	Yes
wiki-forge query "question"	Q&A against wiki	Yes
wiki-forge serve	Start Web UI + API server	No
wiki-forge mcp	Start MCP server (stdio)	No
wiki-forge list	List registered vaults	No
wiki-forge register	Register vault in global registry	No
wiki-forge ingest doc.md	Ingest a document	Yes
wiki-forge lint	Check wiki health (broken links, orphans)	No
wiki-forge drift ./src	Detect code ↔ wiki contradictions	Partial
wiki-forge unify -p a:./a -p b:./b	Merge multi-repo wikis	Yes

---

Code Intelligence (CGC)

Wiki-Forge uses CodeGraphContext (MIT) for code analysis:

• 19 languages: Python, TypeScript, JavaScript, Rust, Go, Java, C/C++, C#, Ruby, PHP, Kotlin, Scala, Swift, Dart, Haskell, Elixir, Perl, and more
• Rust-accelerated: Tree-sitter parsing via Rust native extension
• Call graph: Function → function call edges (1,000-20,000+ per project)
• Operational params: Auto-extracts cron schedules, timeouts, config values

Performance

Repo	Files	Parse	Write	Resolve	Total	Edges
Python backend (300 files)	300	1.0s	7.3s	3.4s	11.6s	11,927
Next.js frontend (300 files)	300	3.0s	4.6s	1.5s	9.1s	5,731
Large TS monorepo (1220 files)	1220	16s	28s	27s	71s	49,958

---

Docker

# Build
docker build -t wiki-forge \
  --build-context cgc=/path/to/CodeGraphContext .

# Run
docker run -d -p 5757:5757 \
  -v ./workspaces:/workspaces \
  -e GEMINI_API_KEY=your-key \
  wiki-forge

Or with docker-compose:

docker-compose up -d
open http://localhost:5757/wiki/

---

MCP Tools (for AI Agents)

Tool	Description	LLM Cost
list_vaults	List all available projects	Free
search_pages	BM25 keyword search	Free
read_page	Read raw wiki page content	Free
get_context	Get ranked pages for a question (AI synthesizes locally)	Free
query_wiki	Full Q&A with LLM synthesis	~$0.001

Claude Desktop Config

{
  "mcpServers": {
    "wiki-forge": {
      "url": "http://your-server:5757/mcp/sse"
    }
  }
}

Claude Code Skill

# Install /wiki slash command globally
curl -sL https://raw.githubusercontent.com/tinhthanh/llm-wiki/main/scripts/install-skills.sh | bash

# Usage in Claude Code
/wiki "how does the auth module work?"

---

Document Converters

Format	Engine	License
PDF	pdf_oxide	MIT/Apache-2.0
DOCX	markitdown	MIT
PPTX	markitdown	MIT
XLSX	markitdown	MIT
HTML	markitdown	MIT
PDF (alternative)	pdfplumber	MIT

All converters are MIT-licensed — safe for commercial SaaS.

---

Architecture

wiki-forge/
├── src/llm_wiki/           ← Python backend
│   ├── cli.py              ← 17 CLI commands
│   ├── server.py            ← FastAPI (REST + SSE + MCP)
│   ├── mcp_server.py        ← MCP tools (5 tools)
│   ├── code/cgc_bridge.py   ← CGC Rust engine bridge
│   └── pipeline/
│       ├── ingest.py        ← Document ingestion (2-step LLM)
│       ├── ingest_code.py   ← Code-to-Doc (5 phases)
│       ├── query.py         ← Q&A (BM25 + graph re-ranking)
│       ├── drift.py         ← Drift detection (7 types)
│       └── lint.py          ← Wiki health check
├── web/                     ← React frontend (Vite + Tailwind)
├── workspaces/              ← Wiki data (markdown flat files)
├── docs/
│   ├── USER_GUIDE.md        ← End-user guide (Dev + BO)
│   ├── DEVELOPER_GUIDE.md   ← Developer guide
│   └── PROPOSAL.md          ← Architecture proposal
├── Dockerfile               ← Multi-stage (Rust + Node + Python)
└── docker-compose.yml       ← One-command deployment

---

Cost Estimate

Operation	LLM Calls	Cost (Gemini Flash Lite)
Index 300 files (code-to-doc)	~25	~$0.04
1 Q&A query	1	~$0.001
Ingest 1 document	2	~$0.002
CGC code index	0	$0 (free)
Search / Browse / Graph	0	$0 (free)

---

Documentation

• USER_GUIDE.md — End-user guide for Dev + BO teams
• DEVELOPER_GUIDE.md — Internal developer guide
• PROPOSAL.md — Architecture & solution proposal

---

License

MIT — free for commercial use.

Dependencies: All production dependencies are MIT/BSD/Apache-2.0 licensed.