LanceDB

LanceDB, made by San Francisco-based LanceDB Inc. after raising $30M in Series A funding, is an AI-native multimodal lakehouse platform that unifies data management for machine learning workflows, handling dataset curation, feature engineering, vector search, and model training from a single table without data duplication or sync jobs. It is built on the open-source Lance columnar file format and offers multimodal data storage supporting text, images, video, and embeddings, vector and full-text hybrid search with SQL filtering at more than 100,000 queries per second, and automated feature engineering with Python UDFs requiring no table rewrites. LanceDB includes data curation tools for deduplication and edge-case detection, git-like versioning with branching, tagging, and rollback, and high-performance training achieving up to 70% Model FLOPS Utilization while scaling to more than 100 billion rows in a single table. It is in production at companies including Netflix, Runway, and ByteDance. Pricing is contact-based for LanceDB Cloud and Enterprise offerings; the underlying Lance file format itself is open source.