Vector DB for AI Agents 'Caliby' Released

Caliby, developed through collaboration by an MIT database research team, is an embedded vector search library designed for AI agent and RAG (Retrieval-Augmented Generation) workflows.

Key features include:

• Overwhelming performance: 4x faster than pgvector, and outperforms FAISS in disk storage scenarios.
• Support for various indexes: Supports HNSW, DiskANN, IVF+PQ indexes, efficiently handling searches across millions to tens of millions of vectors.
• Hybrid storage: Adopts a native hybrid storage approach that supports both text and vectors simultaneously, enhancing agent utility.
• Ease of use: Built on a C++ core with Python bindings, ready to use immediately with just pip install caliby.

Caliby supports CPU and SIMD (AVX-512/AVX2/SSE) acceleration, and aims to be a lightweight engine that runs in-process without any separate external components.