Claude Code Vector Memory

Semantic Memory System for Claude Code

Give Claude persistent memory across conversations by indexing and searching your session summaries.

🚀 Quick Start

Setup (One Time)

The setup automatically:
✅ Creates Python environment
✅ Installs dependencies
✅ Sets up Claude Code integration
✅ Creates global search command
✅ Configures everything for your OS

Start Using

Add Your Summaries

1. Place summary files in claude_summaries/
2. Run: python reindex.py

Features

- Semantic Search: Uses sentence transformers to find conceptually similar past sessions
- Hybrid Scoring: Combines semantic similarity (70%), recency (20%), and complexity (10%)
- Rich Metadata: Extracts titles, dates, technologies, file paths, and more
- ChromaDB Backend: Fast vector similarity search with persistent storage
- Beautiful CLI: Rich terminal output with tables and formatted results
- Cross-Platform: Works on Linux, macOS, and Windows
- Claude Code Integration: Automatic memory search before every task

How It Works

1. Index: Scans your Claude summaries and creates semantic embeddings
2. Search: Finds similar past sessions using vector similarity
3. Integrate: Claude automatically searches memories before each task
4. Learn: Builds on past solutions and approaches

Project Structure

Requirements

- Python 3.8+
- Claude Code
- Your Claude session summaries in ~/.claude/compacted-summaries/

Troubleshooting

- Permission denied? Run: chmod +x *.sh (Linux/macOS)
- Command not found? Restart your shell after setup
- No results? Make sure you've added summaries and run python reindex.py

For more help, see docs/TROUBLESHOOTING.md

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT License - see LICENSE file for details.