LanceDB

LanceDB is an open-source, AI-native multimodal lakehouse that unifies vector, full-text, and hybrid search with data curation, feature engineering, versioning, and high-throughput training-data access. It is built for ML and AI engineering teams running production retrieval and training pipelines at scale.

LanceDB is an AI-native data platform built on the open-source Lance columnar format that unifies data management across the entire ML lifecycle — from raw multimodal files (text, images, video, audio) through feature engineering to model training, retrieval, and deployment. It provides vector, full-text, and hybrid search with SQL filtering, automatic data versioning (branching, tagging, and rollback), deduplication and curation, Python-UDF feature engineering with schema evolution that avoids table rewrites, and high-throughput training-data access (up to ~70% Model FLOPS Utilization with fast random access and no egress bottlenecks). LanceDB scales to 100B+ rows per table and 100K+ queries per second on object storage, and is used by teams at Netflix, Runway, Midjourney, ByteDance, Uber, Character.AI, WeRide, and CodeRabbit. The company raised a $30M Series A led by Theory Ventures in June 2025 (with CRV, Y Combinator, and Databricks Ventures participating) and entered 2026 with Lance-native SQL retrieval via DuckDB, multi-bucket object storage, and continued open-source momentum. It is offered as free Apache-2.0 open source, a usage-based LanceDB Cloud (public beta), and an annual-commit LanceDB Enterprise tier.

Category

Data & Analytics

Pricing

Free, Usage-based, Contact for pricing

Target Market

CTOs, Data Scientists, Enterprise Developers, AI Engineers

Founded

2021

Headquarters

San Francisco, USA

• ✓
  Multimodal search

  Vector, full-text, and hybrid search with SQL filtering across text, images, video, and audio.

• ✓
  Lance open format

  Built on the open-source Lance columnar format designed for fast random access and no-egress training workflows.

• ✓
  Data curation & versioning

  Deduplication, edge-case discovery, and automatic branching, tagging, and rollback of datasets.

• ✓
  Feature engineering

  Python UDFs with automatic updates and schema evolution without rewriting tables.

• ✓
  Training-data access

  High Model FLOPS Utilization with fast random access and no egress bottlenecks for model training.

• ✓
  Scale

  Handles 100B+ rows per table and 100K+ queries per second on object storage.

• •
  Retrieval for RAG and agents

  Serve low-latency vector and hybrid retrieval for generative-AI and agent applications.

• •
  Training-data lakehouse

  Curate, version, and stream multimodal datasets directly into model-training pipelines.

• •
  Multimodal data curation

  Deduplicate and organize embeddings, images, documents, and video for production ML.

Apache-2.0 open source is free and self-hosted; LanceDB Cloud is usage-based (public beta); Enterprise is an annual commit via AWS Marketplace or custom negotiation.

This page was written from 2 sources.

1. 1.lancedb.com — lancedb.comvendor
2. 2.lancedb.com — series a fundingvendor