About LanceDB
LanceDB is an open source database for multimodal AI data. It keeps images, video, text, metadata and embeddings together in one table, so teams can search and prepare training data without many separate systems.
What it does
LanceDB stores data as tables that can hold many kinds of content. You can search across embeddings and metadata, curate training subsets and add new columns computed by Python functions. Dataset branches let an agent or engineer experiment on a copy of the data and review changes before they are merged. Training data can be streamed to GPUs, and the vendor describes global shuffles for faster convergence.
Key features
- Images, video, text and embeddings in one table
- Python functions that add computed columns
- Dataset branches for safe experiments
- Streaming of training data to GPUs
- Search that combines metadata and embeddings
Who it's for
The database suits ML engineers, data teams and research groups who work with large multimodal datasets. A small project with simple text search may find it more than it needs.
Pricing and good to know
The core project is open source under the Apache-2.0 license. The site also offers larger managed options that are arranged through a demo request. Large datasets need storage and compute planning, and the cost depends on the scale of the work.
Highlights
- Images, video, text and embeddings in one table
- Python functions that add computed columns
- Dataset branches for safe experiments
- Streaming of training data to GPUs
- Search that combines metadata and embeddings
