RepoWiki - LightRAG Wiki Generator

Generate comprehensive hierarchical wiki documentation from any code repository using LightRAG knowledge graphs.

Quick Start

Generate Wiki from Current Repository

bash

# Test setup
uv run {baseDir}/scripts/repowiki.py test

# Index and generate wiki (all-in-one)
uv run {baseDir}/scripts/repowiki.py all --extended

# Basic wiki (faster, ~13 pages)
uv run {baseDir}/scripts/repowiki.py all

Index Specific Repository

bash

uv run {baseDir}/scripts/repowiki.py index --repo /path/to/project

Generate from Existing Index

bash

uv run {baseDir}/scripts/repowiki.py generate --extended

Features

✅ Works with any repository - Not limited to specific projects
✅ Auto-detects repo name - From git remote or directory name
✅ Works out of the box - Uses GitHub Copilot models by default
✅ Maximum parallel processing - Optimized for GitHub Copilot Business
✅ Hierarchical organization - 3-4 level deep structure
✅ Smart query modes - global, local, mix, hybrid, naive
✅ Breadcrumb navigation - Easy to navigate
✅ Category indexes - Table of contents for each section

Commands

Test Setup

bash

uv run {baseDir}/scripts/repowiki.py test

Validates configuration, checks dependencies, and verifies repository access.

Index Repository

bash

# Index current directory
uv run {baseDir}/scripts/repowiki.py index

# Index specific repository
uv run {baseDir}/scripts/repowiki.py index --repo /path/to/project

# Custom working directory
uv run {baseDir}/scripts/repowiki.py index --working-dir ./storage

Generate Wiki

bash

# Base wiki (~13 pages, faster)
uv run {baseDir}/scripts/repowiki.py generate

# Extended wiki (~19 pages, comprehensive)
uv run {baseDir}/scripts/repowiki.py generate --extended

# Custom model
uv run {baseDir}/scripts/repowiki.py generate --model gpt-4o

# Custom output directory
uv run {baseDir}/scripts/repowiki.py generate --output ./wiki

All-in-One (Index + Generate)

bash

# Base wiki
uv run {baseDir}/scripts/repowiki.py all

# Extended wiki (recommended)
uv run {baseDir}/scripts/repowiki.py all --extended

# Specific repository
uv run {baseDir}/scripts/repowiki.py all --repo /path/to/project --extended

Wiki Structure

Base Wiki (~13 pages)

code

wiki_docs/
├── README.md                    # Home page
└── 01-overview/                 # Overview & architecture
    ├── README.md
    ├── project-overview.md
    ├── architecture.md
    └── design-decisions.md

Extended Wiki (~19 pages)

code

wiki_docs/
├── README.md                    # Home page
├── 01-overview/                 # Overview & architecture
├── 02-getting-started/          # Installation & configuration
├── 03-core-concepts/            # Key components & workflows
├── 04-api-reference/            # Public API & examples
└── 05-development/              # Dependencies, testing, extensions

Configuration

Environment Variables (Optional)

bash

export REPO_PATH="/path/to/project"
export WORKING_DIR="./repowiki_storage"
export OUTPUT_DIR="./wiki_docs"
export REPO_NAME="My Project"
export LLM_MODEL="github_copilot/gpt-4o"
export EMBEDDING_MODEL="github_copilot/text-embedding-3-small"

Default Configuration

Uses GitHub Copilot models by default (free with GitHub Copilot license):

•LLM Model: github_copilot/gpt-4o (128K context)
•Embedding Model: github_copilot/text-embedding-3-small
•API Key: oauth2 (automatic with GitHub Copilot)
•Working Directory: ./repowiki_storage
•Output Directory: ./wiki_docs

Custom Model Configuration

bash

# Use different model
uv run {baseDir}/scripts/repowiki.py generate --model gpt-4o-mini

# Or set environment variable
export LLM_MODEL="gpt-4o-mini"
uv run {baseDir}/scripts/repowiki.py generate

Query Modes

The knowledge graph supports multiple query modes:

•global - Search across entire codebase
•local - Focus on specific components
•mix - Combine global and local context
•hybrid - Balance breadth and depth
•naive - Simple keyword search

The generator automatically selects appropriate modes for different sections.

Performance

Mode	Pages	Time	Cost
Base	~13	2-3 min	FREE
Extended	~19	5-10 min	FREE

Indexing: First-time indexing may take longer for large repositories
Generation: ~30 seconds with warm cache
Parallelism: Optimized for GitHub Copilot Business (48/96/48 concurrent calls)

Examples

Document Your Own Project

bash

cd /path/to/your/project
uv run /path/to/lightrag-apps/scripts/repowiki.py all --extended

Document Open Source Project

bash

git clone https://github.com/user/project
cd project
uv run /path/to/lightrag-apps/scripts/repowiki.py all --extended

Re-generate After Code Changes

bash

# Re-index updated files
uv run {baseDir}/scripts/repowiki.py index

# Generate fresh wiki
uv run {baseDir}/scripts/repowiki.py generate --extended

File Support

By default, indexes:

•Python files: .py
•Markdown files: .md
•Text files: .txt

Skips files smaller than 50 bytes (configurable via MIN_FILE_SIZE environment variable).

Troubleshooting

Check Setup

bash

uv run {baseDir}/scripts/repowiki.py test

Common Issues

Repository not found

bash

# Specify path explicitly
uv run {baseDir}/scripts/repowiki.py index --repo /full/path/to/project

Import errors

bash

# Reinstall dependencies
uv pip install lightrag-hku openai tiktoken numpy networkx

GitHub Copilot not working

•Ensure you have an active GitHub Copilot license
•Check that you're signed in to GitHub in your IDE
•Try using a different model: --model gpt-4o-mini

Output Files

After generation, you'll find:

code

repowiki_storage/     # Knowledge graph storage (internal)
  └── main/           # Default workspace
      ├── kv_store_*.json
      ├── graph_chunk_*.json
      └── ...

wiki_docs/            # Generated wiki documentation
  ├── README.md       # Start here!
  ├── 01-overview/
  ├── 02-getting-started/
  └── ...

Advanced Usage

Custom Parallel Processing

bash

export MAX_PARALLEL_INSERT=48
export LLM_MODEL_MAX_ASYNC=96
export EMBEDDING_FUNC_MAX_ASYNC=48
uv run {baseDir}/scripts/repowiki.py all --extended

Custom File Extensions

Edit the script's code_extensions configuration to include additional file types.

Multiple Workspaces

bash

# Different workspace for experimental features
export WORKSPACE="experimental"
uv run {baseDir}/scripts/repowiki.py index --repo /path/to/project

Integration

Git Hooks

Add to .git/hooks/post-commit:

bash

#!/bin/bash
uv run /path/to/lightrag-apps/scripts/repowiki.py all --extended

CI/CD Pipeline

yaml

- name: Generate Wiki
  run: |
    pip install repowiki
    repowiki all --extended
    
- name: Deploy Wiki
  uses: peaceiris/actions-gh-pages@v3
  with:
    github_token: ${{ secrets.GITHUB_TOKEN }}
    publish_dir: ./wiki_docs

Technical Details

Built with:

•LightRAG - Knowledge graph framework
•GitHub Copilot models - LLM and embeddings
•NetworkX - Graph operations
•Nano-VectorDB - Vector storage

Architecture:

•Indexer - Scans repository, builds knowledge graph
•Generator - Queries graph, generates hierarchical documentation
•Knowledge Graph - Stores entities, relationships, and context

References

See references/ directory for additional documentation:

•Query modes and strategies
•Performance optimization
•Prompt customization
•Knowledge graph structure

Related Skills

•pptx-creator - Generate presentations from wiki content
•github-pr - Create PRs with wiki updates
•excel - Export wiki metrics to spreadsheets