rustclaw

Memory search engine with hybrid vector/keyword search, MMR re-ranking, and temporal decay.

Rust port of OpenClaw's memory system.

• Hybrid Search - Combines FTS5 keyword search (BM25) with vector similarity search
• HNSW Index - Approximate nearest neighbor search for fast vector queries
• MMR Re-ranking - Maximal Marginal Relevance for diversity-aware results
• Temporal Decay - Exponential time decay with configurable half-life
• SIMD Cosine Similarity - Vectorized similarity computation via wide
• Incremental Embedding - Only embeds new/changed chunks
• Embedding API - OpenAI and Gemini embedding integration
• Session Indexing - JSONL conversation log parsing and search
• MCP Server - Model Context Protocol server for Claude Code integration
• Web UI - Browser-based search interface with live results
• File Watcher - Auto-sync on memory file changes via notify
• SQLite + FTS5 - Persistent storage with full-text search

cargo install --path .

# Show index status
rustclaw -w /path/to/workspace status

# Sync memory files into the index
rustclaw -w /path/to/workspace sync

# Search (keyword-only)
rustclaw -w /path/to/workspace search "rust programming"

# Search with embedding (requires OPENAI_API_KEY)
rustclaw -w /path/to/workspace search "rust programming" --embed

# Generate embeddings for all chunks
rustclaw -w /path/to/workspace embed --provider openai

# Watch for file changes
rustclaw -w /path/to/workspace watch

rustclaw -w /path/to/workspace serve --port 3179
# Open http://127.0.0.1:3179

Add to your Claude Code MCP config (~/.claude/settings.json):

{
  "mcpServers": {
    "rustclaw": {
      "command": "rustclaw",
      "args": ["-w", "/path/to/workspace", "mcp"]
    }
  }
}

Available tools: -