Semantic Git Blame Tool

A powerful semantic search tool for git repositories using LanceDB. Search your git history, find code authors, and understand code evolution using natural language queries.

Features

• 🔍 Semantic Search: Search commits, code changes, and blame data using natural language
• 👤 Author Attribution: Find who implemented specific features or functionality
• 📈 Code Evolution: Track how code has changed over time
• ⚡ Fast Indexing: Efficient vector indexing with LanceDB
• 🔌 IDE Integration: Works with Continue for in-editor code intelligence
• 🌐 HTTP API: REST endpoint for integration with other tools

Quick Start

Prerequisites

• Python 3.9+
• Git
• uv (recommended) or pip

Installation

Option 1: Global Installation (Recommended)

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone the repository
git clone https://github.com/yourusername/git_blame_search.git
cd git_blame_search

# Install globally from local directory
uv tool install .

Option 2: Local Development

# Clone the repository
git clone https://github.com/yourusername/git_blame_search.git
cd git_blame_search

# Install dependencies
uv sync

Usage

1. Index Your Repository

# Index current repository (recent 100 commits)
git_blame_search index --max-commits 100

# Index a specific repository
git_blame_search index --repo /path/to/repo --max-commits 500

# For testing with minimal commits (e.g., new repos)
git_blame_search index --max-commits 1

2. Index Specific Files for Blame Analysis

# Index blame data for important files
git_blame_search index-file main.py
git_blame_search index-file src/core/auth.py

3. Search Commits

# For testing in a new (this) repository
git_blame_search search "init"

# Search by natural language
git_blame_search search "authentication implementation"
git_blame_search search "bug fix for memory leak"
git_blame_search search "refactoring database layer" --limit 20

4. Semantic Blame Search

# Find who wrote specific functionality
git_blame_search blame "error handling logic"
git_blame_search blame "database connection" --file src/db.py

5. Author Attribution

# Find who implemented features
git_blame_search who "authentication system"
git_blame_search who "caching implementation"

Continue Integration

1. Index Your Repository First

# Index your repository before using MCP. It will take a minute on mature projects.
git_blame_search index --max-commits 100

# Optionally index specific files for blame analysis
git_blame_search index-file src/main.py

2. Configure Continue

Add to your ~/.continue/config.yml:

mcpServers:
  - name: git_blame_search
    command: git_blame_search
    args:
      - mcp
    cwd: "."

3. Use in VSCode

• Use git_blame_search in Continue chat
• Ask questions like "who implemented the auth system?"

Example Queries

Finding Code Authors

uv run python git_blame_tool.py who "security vulnerability fix"
uv run python git_blame_tool.py who "REST API implementation"

Understanding Code Evolution

uv run python git_blame_tool.py search "added try-catch blocks"
uv run python git_blame_tool.py search "refactored to use async/await"

Blame Analysis

uv run python git_blame_tool.py blame "validate.*email"
uv run python git_blame_tool.py blame "test_.*authentication" --file tests/

Architecture

The tool uses:

• LanceDB: Vector database for storing and searching embeddings
• Sentence Transformers: Microsoft's CodeBERT for code-aware embeddings
• GitPython: For parsing git repository data
• Rich: For beautiful CLI output

Data Schema

Commits Table

• Commit metadata (hash, author, timestamp, message)
• Files changed, additions, deletions
• Message vector embeddings

Blame Table

• File path, line number, commit hash
• Author information and timestamps
• Code content and vector embeddings

Diffs Table

• Change type, file paths
• Old and new content
• Diff chunk vector embeddings

Performance Tips

Large Repositories

For repositories with 10k+ commits:

# Process in batches
uv run python git_blame_tool.py index --max-commits 1000 --skip 0
uv run python git_blame_tool.py index --max-commits 1000 --skip 1000

Memory Optimization

# Use smaller batches
uv run python git_blame_tool.py index --max-commits 50

# Increase PyTorch memory
export PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512

Faster Embeddings

Modify the embedder for speed:

# In git_blame_tool.py, change:
self.embedder = SentenceTransformer('all-MiniLM-L6-v2')  # Faster

API Reference

CLI Commands

• index - Index a git repository

  • --repo, -r - Repository path (default: current directory)
  • --max-commits, -m - Maximum commits to index

• index-file - Index blame data for a specific file

  • FILE_PATH - Path to file to index

• search - Search commits

  • QUERY - Natural language search query
  • --limit, -l - Number of results (default: 10)

• blame - Search blame data

  • QUERY - Natural language search query
  • --file, -f - Filter by file path
  • --limit, -l - Number of results (default: 5)

• who - Find code authors

  • QUERY - Natural language search query
  • --limit, -l - Number of results (default: 5)

• serve - Start HTTP server for IDE integration

HTTP API

POST /retrieve

{
  "query": "authentication implementation"
}

Response:

{
  "results": [
    {
      "title": "Commit: Add JWT authentication",
      "content": "Author: Jane Smith\nDate: 2024-01-15\n\nAdd JWT authentication middleware",
      "metadata": {
        "type": "commit",
        "hash": "a1b2c3d4e5f6"
      }
    }
  ]
}

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License - see LICENSE file for details

Acknowledgments

• Built with LanceDB - The multimodal vector database
• Uses Microsoft CodeBERT for code-aware embeddings
• Integrated with Continue for IDE support