Summary

sqlite-lembed is an SQLite extension for generating text embeddings using GGUF models via llama.cpp. It targets developers and researchers aiming to integrate AI-powered semantic capabilities directly into SQLite applications, offering efficient, local embedding generation and semantic search without external services.

How It Works

This extension integrates llama.cpp into SQLite, enabling it to load and execute text embedding models in the GGUF format. Users register GGUF models by inserting their path into the temp.lembed_models virtual table. The lembed() SQL function then computes embeddings for input text using these registered models. This approach facilitates local AI-driven text analysis, with embeddings generated in a BLOB format compatible with the sqlite-vec extension for subsequent vector search.

Quick Start & Requirements

Load the extension using .load ./lembed0. GGUF format embedding models are required; a sample all-MiniLM-L6-v2.e4ce9877.q8_0.gguf is available for download. Register models via SQL: INSERT INTO temp.lembed_models(name, model) select 'model-name', lembed_model_from_file('path/to/model.gguf');. Use lembed('model-name', 'your text') to generate embeddings. Pre-converted GGUF models for nomic-embed-text-v1.5 and mxbai-embed-large-v1 are also provided.

Highlighted Details

• Seamless integration with sqlite-vec for vector storage and similarity search.
• Embeddings generated by lembed() use a BLOB format natively understood by sqlite-vec, simplifying data management.
• Includes a practical example demonstrating "semantic search" by embedding news headlines and querying for related articles.

Maintenance & Community

Marked as "A work-in-progress!". Issue #2 requests batch processing support. No specific community channels or detailed contributor information are provided in the README.

Licensing & Compatibility

The README does not specify a software license, preventing determination of commercial use or closed-source linking compatibility without further clarification.

Limitations & Caveats

Key limitations include the lack of batch support for embedding generation, processing each text input individually. Pre-compiled versions lack GPU acceleration, potentially leading to slow embedding generation. Users requiring faster performance should compile the extension from source.