Open-source AI-native multimodal lakehouse unifying data curation, feature engineering, vector search, and model training at petabyte scale.
LanceDB is an AI-native multimodal lakehouse and vector database built around the Lance columnar data format. It provides a unified platform for managing training data from raw files to production-ready features, supporting vector/semantic, full-text, and hybrid search at up to 100K QPS. Features include declarative feature pipelines, automatic embedding updates, versioning and branching without data duplication, and GPU-friendly storage for efficient training. LanceDB is open source and powers memory for CrewAI across 6 million monthly downloads. LanceDB is a product of LanceDB.
For people
For agents