feat(reasoningbank): Local embeddings with transformers.js - no API key required

[ruvnet]

Summary

Implements local embedding generation using transformers.js, eliminating API key requirements and enabling completely offline semantic search.

Changes

Core Implementation (v1.8.6-1.8.10)

Local Transformer Model:

• Uses @xenova/transformers with Xenova/all-MiniLM-L6-v2
• 384-dimensional embeddings (optimized from 1024)
• Completely free and offline after initial 23MB model download
• 50-100ms per embedding with LRU caching

NPX Environment Detection:

• Automatically detects npx execution (npm_lifecycle_event === 'npx')
• Skips transformer initialization in npx (avoids ONNX/WASM errors)
• Shows helpful message for semantic search via global install
• Can force transformers with FORCE_TRANSFORMERS=true

Files Modified:

• src/reasoningbank/utils/embeddings.ts - Complete rewrite with transformers.js
• src/reasoningbank/config/reasoningbank.yaml - Updated to local provider
• package.json - Added @xenova/transformers, updated build
• .npmignore - Excluded Rust artifacts (package: 166MB → 1.6MB)

Features

✅ WASM backend configuration for Node.js compatibility
✅ NPX environment detection (clean UX, no confusing errors)
✅ LRU cache (1000 entries) with TTL expiration
✅ Graceful fallback to hash-based embeddings
✅ Lazy initialization (no startup delay)

Configuration

embeddings:
  provider: "local"
  model: "Xenova/all-MiniLM-L6-v2"
  dimensions: 384
  cache_ttl_seconds: 3600

Usage

# NPX (hash-based fallback - clean experience)
npx claude-flow@alpha memory store "key" "value"
# Output: [Embeddings] NPX environment detected - using hash-based embeddings

# Global install (true semantic search)
npm install -g claude-flow
claude-flow memory store "key" "value"
# Output: [Embeddings] Local model ready! (384 dimensions)

Benefits

• 🆓 Zero Cost: No API fees
• 🔒 Privacy: All processing local
• 📴 Offline: Works without internet (after download)
• ⚡ Fast: 50-100ms inference
• 💾 Efficient: Smart caching
• 🛡️ Reliable: Automatic fallback

Testing

• ✅ Model initialization and download
• ✅ Embedding generation (384 dims)
• ✅ NPX detection and clean fallback
• ✅ Cache functionality
• ✅ Semantic similarity search

Version History

• v1.8.6: Initial local embeddings
• v1.8.7: Config file copying, WASM fixes
• v1.8.8: WASM backend proxy config
• v1.8.9: NPX detection (user_agent)
• v1.8.10: Improved NPX detection (lifecycle_event) ✅

Closes #38