LanceDB

LanceDB is a database for AI workloads that the vendor describes as a multimodal lakehouse. It stores vectors alongside the raw data they describe, such as text, images, video and robotics data, so that search, feature engineering and model training can run against one table rather than several systems. It is built on the Lance data format, which is designed for fast random access and for keeping data in object storage. Published material covers vector search that scales from millions to billions of rows, with search precision tunable at query time, hybrid scans, reranking, native multimodal SQL, and distributed search benchmarked across 10 billion vectors. Documented use cases include LLM pretraining pipelines, vision language model fine tuning, agent memory and mining fleet data for robotics and autonomous systems. The blog describes LeRobot training directly from Lance datasets and CrewAI using LanceDB for memory. The tool is aimed at machine learning engineers, AI application developers and data platform teams, with enterprise offerings for larger deployments. The company announced a $30 million Series A in 2025. Documentation and community channels are linked from the site. The site does not publish plan prices. Its pricing page is a contact form, and the vendor directs buyers to get in touch.