zvec: Lightweight Local Vector Search for Your Own Knowledge Base

Find: a lightweight local vector search that lets you quickly build your own knowledge base.

Alibaba’s open-source project zvec has already gained 10.3k stars on GitHub and is actively used inside Alibaba in large production scenarios.

Usually, for vector search you have to spin up a separate service: Pinecone, Weaviate, Qdrant, or something else. On top of that—setup, infrastructure, and deployment. For small projects, this often turns out to be too hardcore.

With zvec, everything is simpler. It embeds the entire vector database directly into your process—roughly the same way SQLite works with relational databases. Install via pip, create a collection with a few lines of code, load the data—and you can run similarity search. No servers, config files, or extra infrastructure; everything is stored in a single local file.

zvec can search among hundreds of millions of vectors in milliseconds and supports hybrid search: vector + full-text. It runs on laptops, servers, and even on edge devices.

The v0.5.0 update from June 12 noticeably expanded the project’s capabilities:

↳ Native full-text search without external search engines
↳ Hybrid search that combines vectors, keywords, and filters in a single query
↳ A new on-disk index that significantly reduces memory usage on large datasets
↳ Zvec Studio — a visual tool for browsing data and debugging without writing code