DATABASES Signal 490
ParqDB eliminates the query server by running vector search in the browser on Parquet over HTTP
↗
Illustration only Photo by Pierre Bamin on Unsplash
ParqDB builds an IVF-LVQ8 index over MiniLM embeddings, stores it as Parquet in object storage, and runs vector search in the browser via HTTP range reads without a query server.
This shifts vector search from a server-side database to a static-file architecture, eliminating query-server infrastructure and keeping user data in the browser. Engineers must pre-build and publish the index, and the browser must handle embedding and ranking, which may limit dataset size and update frequency.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
ParqDB uses an IVF-LVQ8 index over MiniLM embeddings, stored as immutable Parquet files in object storage.
The browser embeds the query, reads only required byte ranges via HTTP range requests, and ranks results in WASM.
No query server is involved, so user data is not sent to a server for search.
THE CLUSTER