LanceDB

@netflix 🗓️ Session details: #agenda" target="_blank" rel="nofollow noopener">dataengineeringopenforum.com/?session=power… 🔗 Register link: luma.com/deof2026?utm_s…

Only 2 weeks away from Data Engineering Open Forum 2026 in SF on April 16! Join us for "Powering Netflix's Multimodal Feature Engineering at Scale" and dive into how @netflix curates multimodal features across large video & image corpora, with LanceDB serving as the core storage and query layer for multimodal data.

TONIGHT! Join us, @anyscalecompute, and @ExaAILabs as we dive into Exa's web-scale AI search stack, powered by Lance and Ray. We'll walk through each layer of the system: • Lei Xu (LanceDB) — multimodal storage and vector retrieval with Lance • Goutam Venkatramanan (Anyscale) — distributed processing and embedding pipelines with Ray • Hubert Yuan (Exa) — how Exa runs AI search on top of this infrastructure If you’re building AI search or large-scale retrieval systems, it should be a fun deep dive into how this actually runs in production: luma.com/3ay9xdwb

Just merged into Lance: a completely new FTS index layout 🚀 9.35x faster index builds (34s → 3.6s) 🚀 8.87x smaller index size (955MB → 107MB) 🚀 3.41x faster phrase queries (21ms → 6ms) Block bitpacking + delta encoding + new phrase query algo. Opt in today with LANCE_FTS_FORMAT_VERSION=2 github.com/lance-format/l…

@openclaw 4/4 Shoutout to @AISuperDomain for creating memory-lancedb-pro! Check out the repo: github.com/CortexReach/me…

@openclaw 3/4 Key difference vs memory-lancedb: top-k retrieval vs larger candidate pool + reranking Full breakdown: lancedb.com/blog/openclaw-…

1/4 If your agent memory requires multiple tool calls to retrieve context, recall quality drops. We benchmarked 3 @OpenClaw memory backends on LOCOMO (long-term memory QA). Same dataset, same chunks. 🧵👇

5/5 These constraints are why Lance Blob V2 introduces multi-semantic blob storage. Different storage semantics can coexist within the same column while the system manages routing and lifecycle. Full blog: lancedb.com/blog/lance-blo…

4/5 Problem 3: Lifecycle governance When blobs live outside table metadata, lifecycle management becomes manual. Tracking references, cleaning up orphaned files, and maintaining consistency across versions becomes operational overhead.

1/5 Multimodal datasets break most blob storage strategies. Why? Because production workloads hit three constraints at the same time: - mixed blob sizes - existing object storage pipelines - lifecycle governance 🧵👇

@huggingface 3/3 Updates are incremental: add new columns (labels, embeddings) without rewriting existing data—only new data is written. Blog: lancedb.com/blog/upload-la…

@huggingface 2/3 Once it’s live, query it directly via hf://—no download required. Run vector search, full-text search, or SQL, and filter on nested fields without flattening.

1/3 Upload a Lance dataset to 🤗 @huggingface in 3 steps: 1. Convert raw data (JSON, images) into a Lance table 2. Add embeddings + indexes 3. Upload the table (data + indexes + versions) huggingface.co/datasets?libra…

Huge shoutout to @OnlyXuanwo for kicking off this work during his Christmas break last year 👏 And thanks to the @duckdb team for pushing this through 🦆

Check out the documentation here: duckdb.org/docs/stable/co…

Excited to announce that Lance is now a 𝗰𝗼𝗿𝗲 𝗲𝘅𝘁𝗲𝗻𝘀𝗶𝗼𝗻 of @duckdb! 🦆🚀 The Lance core extension allows DuckDB users to read and write Lance tables, enabling advanced operations directly within DuckDB, including vector similarity search, full-text search, and hybrid search across local or S3-based datasets.

"just worked" <3