v0.9.20: OSS Embedding Support

🚀 weave-mcp v0.9.20: OSS Embedding Support

This release integrates weave-cli v0.9.20 features, focusing on Phase 0: OSS Embedding Support - enabling cost-effective, local embedding generation with no API keys required.

✨ New Features

🎯 3 New MCP Tools

1. list_embedding_models

Discover all available embedding models with metadata:

• 16+ models across 5 providers:
  • OpenAI (text-embedding-3-small, text-embedding-3-large, text-embedding-ada-002)
  • Ollama (nomic-embed-text, mxbai-embed-large, snowflake-arctic-embed)
  • sentence-transformers (all-mpnet-base-v2, all-minilm-l6-v2, multi-qa-mpnet-base-dot-v1, and more)
  • Cohere (embed-english-v3.0, embed-multilingual-v3.0)
  • Voyage (voyage-2, voyage-large-2)
• Filter by provider or OSS status
• Shows dimensions, max tokens, and provider info

2. reembed_collection

Switch embedding models 20x faster than re-ingestion:

• Re-generates embeddings from existing document chunks
• No need to re-ingest documents
• Supports batch processing for efficiency
• Works with all supported providers (OpenAI, Ollama, sentence-transformers)
• Validated: +11% quality improvement, $240/year savings per million docs

3. check_embedding_provider_availability

Verify provider configuration before use:

• Checks if API keys are configured (OpenAI)
• Tests local services (Ollama, sentence-transformers)
• Validates connectivity and authentication

🎁 Key Benefits

• 💰 Cost Savings: Use free OSS models instead of paid APIs
• 🔒 Privacy: 100% local embedding generation (no data sent to third parties)
• ⚡ Performance: 20x faster model switching via re-embedding
• 🎯 Quality: Choose the best model for your use case
• 🔄 Flexibility: Easy to experiment with different models

📦 Dependencies

• Updated weave-cli from v0.9.15 to v0.9.20
• Added support for:
  • embeddings package (model registry with auto-dimension detection)
  • reembedding package (efficient embedding regeneration)
  • reembedding/providers package (OpenAI, Ollama, sentence-transformers)

🔧 Technical Details

• Model Registry: Auto-detects dimensions for 16+ models
• Provider Factory: Unified interface for all embedding providers
• Batch Processing: Efficient re-embedding with configurable batch sizes
• Error Handling: Graceful fallbacks and detailed error messages
• Metrics: All new tools instrumented with Prometheus metrics

📊 Production Validation

Results from Client0 validation (documented in weave-cli):

• +11% quality improvement with sentence-transformers vs OpenAI
• $240/year savings per million documents
• 20x faster model switching (minutes vs hours)

🚀 Usage Examples

List all OSS models:

```json
{
"tool": "list_embedding_models",
"arguments": {
"oss_only": true
}
}
```

Re-embed a collection:

```json
{
"tool": "reembed_collection",
"arguments": {
"collection": "articles",
"new_model": "sentence-transformers/all-mpnet-base-v2",
"batch_size": 100
}
}
```

Check if Ollama is available:

```json
{
"tool": "check_embedding_provider_availability",
"arguments": {
"provider": "ollama"
}
}
```

📝 What's Next (Optional Enhancements)

Future releases may include:

• Update create_collection to support embedding_model parameter
• Auto-detect collection embedding model in query/search tools

🙏 Credits

Built with weave-cli v0.9.20

---

🤖 Generated with Claude Code

Co-Authored-By: Claude noreply@anthropic.com