LanceDB

What it does

LanceDB is a multimodal lakehouse for AI data. It covers curation, feature engineering with Python UDFs, and unified vector, full-text, and hybrid search with SQL filters. Teams can train directly from curated data with up to 70% MFU and no data movement bottlenecks.

LanceDB is a data platform for AI teams that need to curate, enrich, and search multimodal data. The homepage lists four main use cases: curation, feature engineering, search and retrieval, and training. Curation covers finding optimal distributions, deduplicating massive datasets, and surfacing edge cases. Feature engineering uses Python UDFs with automatic updates and no table rewrites. Search combines vector, full-text, and hybrid retrieval with SQL filters. Training runs directly from curated data with up to 70% MFU and no data movement bottlenecks. Blog posts describe work such as a GPT-2-like pretraining pipeline, a 10B vector search setup, and an OCR pipeline for handwritten notes. The site does not state pricing, a free plan, or open source licensing on the pages reviewed, so those details remain unknown. A reader should check the docs and enterprise pages for limits and terms.