LanceDB

LanceDB is an open-source, serverless vector database for developers and teams building RAG pipelines, LLM agents, recommendation systems, and hybrid search over multimodal data.

Key capabilities include:

• Multimodal storage in one table: stores embeddings, raw data, and metadata together (including vectors, text, images, documents, videos) on S3-compatible object storage.
• Fast similarity and hybrid retrieval: supports approximate nearest neighbor (ANN) search with IVF_PQ indexing, plus hybrid search that combines semantic retrieval with keyword-style querying and metadata filtering.
• SQL-compatible analytics and tooling interoperability: offers SQL querying, integrates with Apache Arrow for zero-copy access, and works with common data tools such as Pandas and Polars.
• Developer-friendly deployment options: supports serverless, stateless compute and embedded database use cases (including browser and edge function deployments), with SDK support for Python and JavaScript.

Pricing: Not specified.

Notable for its Lance columnar format (Rust-based) optimized for ML workloads, plus native versioning and object-storage-first design that avoids separate database infrastructure for storing embeddings and metadata together at scale.