claude-smart-fork 0.1.0

Semantic search across Claude Code sessions to find the best context for new tasks

claude-smart-fork

Semantic search across your Claude Code sessions. Find the most relevant context for new tasks instead of starting from scratch.

The Problem

You've had hundreds of Claude Code sessions. Each one contains valuable context: decisions made, patterns established, problems solved. But when you start a new task, all that knowledge is locked away in session files you can't easily search.

The Solution

claude-smart-fork indexes your Claude Code session history and lets you semantically search for relevant past sessions. Found a good match? Fork from that session and continue with all the context already loaded.

$ smart-fork search "add rate limiting to Express API"

🔍 Top Matching Sessions:

1. 🟢 [94%] a1b2c3d4...
   📁 ~/projects/my-api
   📝 API middleware and request handling patterns
   🛠️ TypeScript, Express, Redis

2. 🟡 [82%] e5f6g7h8...
   📁 ~/projects/rate-limiter
   📝 Redis-based rate limiting implementation

To fork the top result:
  claude --resume a1b2c3d4-full-session-id

Installation

Minimal Install (Keyword Search)

pip install claude-smart-fork

This gives you the CLI with basic keyword-based search. No heavy dependencies.

With Vector Search (Recommended)

pip install "claude-smart-fork[chromadb,embeddings]"

Adds semantic search using local embeddings (~300MB for the model, downloaded once).

With LLM Summarization

# Using Claude API
pip install "claude-smart-fork[claude]"

# Using local Ollama
pip install "claude-smart-fork[ollama]"

Everything

pip install "claude-smart-fork[all]"

Quick Start

1. Initialize

smart-fork init

This creates ~/.claude-smart-fork/ with default configuration.

2. Index Your Sessions

# Preview what would be indexed
smart-fork index --dry-run

# Index all sessions
smart-fork index

3. Search

smart-fork search "implement OAuth authentication"

4. Fork

Copy the session ID from the search results and use Claude Code's built-in resume:

claude --resume <session-id>

Configuration

Configuration is stored in ~/.claude-smart-fork/config.json:

{
  "sessions_path": "~/.claude/projects",
  "backend": "chromadb",
  "embedding_model": "nomic-ai/nomic-embed-text-v1",
  "summarizer": "simple",
  "auto_index": true,
  "search_results_limit": 5
}

Backend Options

Backend	Install	Description
sqlite	Base	Keyword search, no vectors
chromadb	[chromadb]	Vector search with local embeddings

Embedding Models

Model	Size	Quality	Install
nomic-ai/nomic-embed-text-v1	270MB	Best for code	[embeddings]
all-MiniLM-L6-v2	80MB	Good, faster	[embeddings]
openai	API	Excellent	Requires API key

Summarizers

Summarizer	Install	Description
simple	Base	Keyword extraction, no LLM
claude	[claude]	Claude API summarization
ollama	[ollama]	Local Ollama models

Claude Code Integration

Automatic Indexing with Hooks

Add to ~/.claude/settings.json:

{
  "hooks": {
    "SessionEnd": [
      {
        "type": "command",
        "command": "smart-fork index-session $CLAUDE_SESSION_ID"
      }
    ]
  }
}

Now sessions are automatically indexed when they end.

Custom Slash Command

Create ~/.claude/commands/detect-fork.md:

# /detect-fork

Search for relevant past sessions to fork from.

## Usage

Run: `smart-fork search "$ARGUMENTS"`

Present the results and offer to provide the fork command.

CLI Reference

# Initialize configuration
smart-fork init [--force]

# Index sessions
smart-fork index [--dry-run] [--limit N] [--since DATE]

# Index a specific session
smart-fork index-session <session-id>

# Search sessions
smart-fork search <query> [--limit N] [--project PATH]

# Show statistics
smart-fork stats

# Show configuration
smart-fork config [--edit]

# Update configuration
smart-fork config set <key> <value>

How It Works

1. Parsing: Reads Claude Code JSONL session files from ~/.claude/projects/
2. Summarization: Extracts key information (topic, decisions, files, technologies)
3. Embedding: Converts summaries to vectors using local models
4. Storage: Stores vectors in ChromaDB for fast similarity search
5. Search: Converts your query to a vector and finds nearest neighbors

Development

# Clone the repo
git clone https://github.com/a-bekheet/claude-smart-fork.git
cd claude-smart-fork

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run linting
ruff check src tests
ruff format src tests

# Type checking
mypy src

Contributing

Contributions are welcome! Please read CONTRIBUTING.md for guidelines.

Roadmap

• Cross-machine sync via Git
• Project-scoped search
• Time-based queries ("what was I working on Tuesday?")
• Session similarity chains
• VS Code extension
• Web UI for browsing sessions

License

MIT License - see LICENSE for details.

Acknowledgments

• Built for use with Claude Code
• Uses ChromaDB for vector storage
• Embeddings via sentence-transformers