LanceDB is a columnar-format vector database built for multimodal AI applications. It stores data in the Lance format (an improved version of Parquet optimized for ML), which lets you efficiently store and query text, images, video, and vectors in a unified schema.
LanceDB runs embedded in your application like Chroma — zero-ops, no server required. Data is persisted to the Lance columnar format, which supports efficient random access, schema evolution, and versioning. You can query via SQL or a Python API. Because it uses Parquet-compatible storage, it interoperates with the broader data stack (Pandas, Polars, DuckDB, Spark).
Open source: free. LanceDB Cloud in private beta, pricing TBD. Self-hosting has no license fees.
AI research teams, data scientists, and multimodal product builders. Particularly popular for image/video search and in data science workflows where interop with Pandas/Polars matters.