Top AI Product

Most vector databases want you to run a server. Zvec, open-sourced by Alibaba’s Tongyi Lab, goes the other way — it embeds directly into your application with no server, no daemon, and no network calls. The team pitches it as “the SQLite of vector databases,” and it is built on Proxima, Alibaba’s production vector search engine.

## What Zvec does

Zvec runs entirely in-process, which makes it aimed squarely at retrieval-augmented generation, semantic search, and agent workloads that need to run locally — on laptops, phones, or other constrained edge hardware. No infrastructure to provision: install it, and your app has vector search. That on-device angle matters as more AI features move off the cloud for latency, cost, and privacy reasons.

## The v0.5.0 update

The June 12 release pushes it past pure vector search. Zvec 0.5.0 adds native full-text search with natural-language or structured queries, hybrid retrieval that combines full-text and vector results, and a new DiskANN index that keeps most of the index on disk to slash memory use on large datasets. That combination — keyword plus semantic, in one embedded engine — is what most RAG stacks normally bolt together from separate tools.