Introducing Lance Community Governance 🛡️
We’ve launched a dedicated Lance discord focused entirely on the format, feature discussions, proposals, and real-world use cases. In addition, Lance now also has its own website and GitHub organization to maintain its own ecosystem for its user community.
- Read the announcement here: https://lancedb.com/blog/lance-community-governance/
- Learn more about the community: https://lance.org/community/
- Join the new Lance discord: https://discord.gg/lance
- Give Lance format a star: https://github.com/lance-format
From BI to AI: A Modern Lakehouse Stack with Lance and Iceberg 🧊
The modern lakehouse stack is composed of six layers. File formats, table formats, and catalog specs are just storage definitions. All compute power actually lives in the object store, catalog services, and compute engines.
What makes Lance different is that it spans all three storage layers at once: file format, table format, and catalog spec. Iceberg operates at two: table format and catalog spec.
Iceberg remains a strong choice for large-scale OLAP and BI workloads. Lance complements it by addressing AI and multimodal data requirements with an Arrow-native layout, high-performance indexing, and built-in interop with Parquet.
Together, both formats can coexist in the same lakehouse stack: Iceberg for BI, Lance for AI.
Netflix’s Multimodal Search Demo 🔍
Here is a demo that Pablo Delgado from Netflix put together for Netflix and LanceDB’s joint talk at Ray Summit 2025 (you can find the session recording below). This video highlights how to search through hundreds of terabytes of multimodal data with negligible latency and perform multimodal data understanding at scale.
The demonstration showcases a sophisticated multimodal embedding system that enables semantic search across Netflix’s vast video catalog. The system supports multiple embedding modes (text-to-text, text-to-image, image-to-image, and image-to-text) allowing researchers to query content using either natural language descriptions or visual references. Each video frame is enriched with metadata that captures not only visual content but also contextual details like scene composition, lighting, mood, and subject matter.
📺 Recordings you missed!
📊 LanceDB Enterprise Product News
We have enabled full-text search in SQL to reach parity with our Python API capabilities. We have also introduced incremental indexing using SPFresh, eliminating the need for full reindexing while maintaining centroid freshness and reducing cold latency significantly.
🫶 Community Contributions
- VoyageAI integration ( documentation , lancedb#2713 )
- Manifest summary support in Java API (AntGroup) ( lance#5092 )
A heartfelt thank you to our community contributors of lance and lancedb this past month:
@mykolaskrynnyk @valkum @fzowl @rm-dr @ozkatz @ddupg @majin1102 @shiyajuan123 @jaystarshot @wojiaodoubao @zhangyue19921010 @fenfeng9 @fangyinc @HaochengLIU @Pmathsun @ztorchan @yanghua @timsaucer @fangbo @steFaiz @niyue @xloya @oceanusxiv @luohao @rahil-c @BorenTsai @Maxwell-Guo @teh-cmc @camilesing
🌟 Open Source Releases Spotlight
| Feature | Version | Description |
|---|---|---|
| LanceDB | 0.22.3 | IVF_RQ index type (
lancedb#2687
) Support creating permutation view for PyTorch Data Loader ( lancedb#2552 ) Add FTS UDTF support ( lancedb#2755 ) Expand support for multivector colpali models ( lancedb#2719 ) |
| Lance | 0.39.0 | Incremental Vector Indexing with SPFresh (
lance#4837
) Dynamic AWS Credentials vending for Lance datasets ( lance#4905 ) |
| Lance Namespace | 0.0.21-0.2.1 | DirectoryNamespace v2 which supports multi-level namespace, with REST namespace adapter in rust, python and java (note that the features have been refactored into the main lance repository and released with the main lance SDKs going forward) ( lance#5292 ) |
