LanceDB

LanceDB is an open source vector database for developers building AI retrieval applications and teams working with training datasets. Its embedded library runs locally or in your own cloud under the Apache 2.0 license. Cloud and Enterprise offerings provide managed infrastructure for production workloads.

Built on the Lance columnar format, it keeps raw content, embeddings and metadata in the same table. That means an application can store images or video alongside their searchable vectors, rather than maintain separate stores for each. Supported data also includes text and point clouds.

Vector similarity, full-text and hybrid search work with SQL filters, so retrieval can combine semantic matches with text queries and metadata constraints. Python, TypeScript and Rust SDKs let developers use the database from their applications. GPU support is available for building vector indexes.

Dataset changes are automatically versioned. Teams can branch an experiment, compare results or roll back without duplicating the data. They can also add feature columns without rewriting the existing table.

The broader platform covers dataset curation, feature engineering and model training from those same tables. Curation includes deduplication and finding edge cases for labeling. Feature pipelines support Python functions and automatic embedding updates, while training storage provides random access and global shuffling to feed GPUs.