zvec.org

Zvec is a lightweight vector database that runs directly inside your application — no separate server or cloud service needed. You install it via pip, define a schema, insert vectors, and start searching. It handles dense and sparse vectors, supports filtered and grouped queries, and claims millisecond search at billion-vector scale. The whole thing is designed to be simple: you don't need to manage infrastructure or learn a complex query language.

What makes Zvec stand out is its in-process architecture. Instead of sending requests over a network, the database lives in your app's memory space, which cuts latency dramatically. The Python API is straightforward — create a collection, insert documents with vector embeddings, and query with a few lines of code. It also supports hybrid search (combining semantic search with metadata filters) and a GROUP BY style clause for grouped similarity searches. Benchmarks on the Cohere 10M vector dataset show over 8,500 queries per second and an index build time of about an hour.

Zvec is built for developers building AI features like retrieval-augmented generation (RAG), image similarity search, or natural language code search. If you're prototyping a chatbot that needs to pull from a knowledge base, or building a visual search engine for product images, Zvec gives you a fast, local vector store without the overhead of a full database cluster. It's especially useful for small teams or solo developers who want to add semantic search without managing another service.