lance

Lance is an open-source lakehouse format designed for multimodal AI workloads. It provides a unified file, table, and catalog specification to build a complete lakehouse on object storage. Key features include expressive hybrid search (combining vector similarity, full-text BM25, and SQL analytics), lightning-fast random access (100x faster than Parquet or Iceberg for random access), native support for multimodal data (images, videos, audio, text, embeddings), efficient data evolution with backfilled columns, zero-copy versioning with ACID transactions and time travel, and rich ecosystem integrations with Apache Arrow, Pandas, Polars, DuckDB, Spark, Ray, Trino, Flink, and open catalogs. The project includes a core Rust implementation, Python bindings (PyO3), Java bindings (JNI), and documentation. Benchmarks show sub-millisecond average response times for vector search on 1M vectors and significant performance gains over Parquet for random access and analytics. Lance is actively developed and welcomes contributions.