LanceDB

LanceDB is an AI-native multimodal lakehouse platform that consolidates scattered training data into a single table for curation, feature engineering, search and retrieval, and model training. It is built on an open source format for multimodal data and is designed to operate at petabyte to exabyte scale, with automatic versioning of writes so datasets can be branched, rolled back, and tagged without duplicating data.

The platform supports vector/semantic, full-text, and hybrid search combined with SQL filters against one table. Feature engineering uses Python UDFs with declarative pipelines that add columns and update embeddings without rewriting existing data. Curation covers deduplication and edge-case identification, while training runs from the same table with dedicated fast storage for random access and shuffles. Search is horizontally scalable across raw bytes, features, metadata, and embeddings.

LanceDB targets teams building and training AI models at large scale, including agent-based retrieval workloads. The homepage directs visitors to contact sales for access, and no pricing details or plan tiers are stated.