lancedb

Why we included this project

LanceDB is an embedded vector store that runs inside your own process, so you get fast similarity search without provisioning a separate database server. It stores vectors and metadata in the Lance columnar format, which also holds raw multimodal data like images, text, video, or point clouds, and you can query the same store by vector similarity, full-text, or SQL. The library works from Python, TypeScript, and Rust, and while it feels like a local tool, the same design scales to cloud and enterprise setups when your data grows. Because storage is zero-copy and versioned, you can iterate on training datasets and feature stores without duplicating infrastructure. If you already use LangChain or LlamaIndex, drop-in connectors keep the switch small.