
LanceDB
Serverless multimodal vector database built on the Lance columnar format.
About
LanceDB is an open-source vector database designed as a 'multimodal lakehouse': it stores vectors alongside text, images, video, and metadata in the Lance columnar format, and serves vector similarity, full-text, and SQL search from one engine. It runs embedded (no server) or at cloud scale, with zero-copy versioning of datasets. Python, TypeScript, and Rust SDKs plus LangChain/LlamaIndex integrations make it a drop-in vector store for RAG and agent pipelines; the company also sells a managed cloud tier.
What it can do
- **Embedded + cloud**: zero-server local mode; managed cloud for scale
- **Multimodal storage**: vectors, text, images, video, point clouds in one table
- **Three query modes**: vector similarity, full-text search, and SQL
- **Dataset versioning**: zero-copy version management without extra infra
- **Lance columnar format**: efficient storage shared with the broader Lance ecosystem
- **Framework integrations**: LangChain, LlamaIndex, Pandas, Polars, DuckDB
Setup
```bash
pip install lancedb
```
```python
import lancedb
db = lancedb.connect("/tmp/db") # embedded, no server
table = db.create_table("docs", data=[{"id": 1, "vector": [0.1, 0.2]}])
result = table.search([0.1, 0.2]).limit(5).to_pandas()
```
See the quickstart at docs.lancedb.com. Managed cloud available at lancedb.com for production scale.
Prerequisites
Python 3.9+ (or Node.js 18+ / Rust for other SDKs). No server or credentials needed for embedded mode.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: 11,652 GitHub stars (log-scaled vs catalog peers); blended with ~7,689,981 monthly package downloads; last push 0d ago; 100 releases/12mo; 745 open issues. Manual review: credible multimodal lakehouse; 7.7M/mo PyPI figure inflated by mirrors/CI; cloud-tier roadmap gravity worth watching; 'billions of vectors' is best-case benchmarking. License: Apache-2.0.
★ 11,652 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.3/10
- Maintenance · recency, releases, issue load
- 8.2/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · 11,652 GitHub stars (log-scaled vs catalog peers)
- · blended with ~7,689,981 monthly package downloads
- · last push 0d ago; 100 releases/12mo; 745 open issues
- · manual review: credible multimodal lakehouse; 7.7M/mo PyPI figure inflated by mirrors/CI; cloud-tier roadmap gravity worth watching; 'billions of vectors' is best-case benchmarking
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.