Skip to main content
Zvec is an open-source, in-process vector database — lightweight, lightning-fast, and designed to embed directly into applications. Built on Proxima (Alibaba’s battle-tested vector search engine), it delivers production-grade, low-latency, scalable similarity search with minimal setup.

​
Why Zvec?

Unlike traditional client-server vector databases, Zvec runs in-process as a library that embeds directly into your application. This architectural choice brings several key advantages:
  • Zero infrastructure overhead — No servers to deploy, configure, or maintain
  • Ultra-low latency — Eliminate network hops and serialization overhead
  • Simplified deployment — Works anywhere your code runs: notebooks, servers, CLI tools, or edge devices
  • No network dependencies — Perfect for offline environments and air-gapped systems
  • Easier debugging — Stack traces and profiling work seamlessly with your application code

​
Key Features

Blazing Fast

Searches billions of vectors in milliseconds, powered by Proxima’s battle-tested algorithms

Simple, Just Works

Install and start searching in seconds. No servers, no config, no fuss

Dense + Sparse Vectors

Work with both dense and sparse embeddings, with native support for multi-vector queries in a single call

Hybrid Search

Combine semantic similarity with structured filters for precise results

​
Dense and Sparse Vector Support

Zvec natively supports both dense and sparse embeddings, enabling flexible multi-vector search strategies:
  • Dense vectors: Traditional embeddings from models like BERT, OpenAI, or Sentence Transformers
  • Sparse vectors: BM25, SPLADE, or other sparse representations for keyword-based semantic search
  • Multi-vector queries: Query multiple vector fields simultaneously in a single call

​
Hybrid Search Capabilities

Combine the power of semantic similarity with structured filtering: 
# Search with semantic vectors AND structured filters
results = collection.query(
    vectors=VectorQuery("embedding", vector=[0.1, 0.2, 0.3, 0.4]),
    filter="category == 'technology' AND publish_date > '2024-01-01'",
    topk=10
)

​
Use Cases

Zvec is ideal for applications that need vector search without the complexity of distributed systems:

RAG (Retrieval-Augmented Generation)

Build intelligent chatbots and AI assistants with semantic document retrieval

Semantic Search

Power search engines that understand meaning, not just keywords

Recommendation Systems

Find similar items, products, or content based on embeddings

Document Clustering

Group similar documents, images, or other content automatically

​
When to Choose Zvec

Choose Zvec when you need:
  • Fast prototyping and development without infrastructure setup
  • Low-latency search within a single application instance
  • Simplified deployment for edge devices or embedded systems
  • Local-first applications with offline support
  • Testing and experimentation before scaling to distributed systems
Consider alternatives when you need:
  • Multi-tenant SaaS applications with millions of users
  • Distributed search across multiple machines
  • High availability with automatic failover
  • Built-in auth, RBAC, and enterprise security features

​
Comparison to Other Vector Databases

FeatureZvec (In-Process)Client-Server DBs
DeploymentEmbed in applicationSeparate server(s)
Network latencyNoneAdds milliseconds per query
InfrastructureNone requiredRequires hosting, monitoring
ScalabilitySingle-processHorizontal scaling
Best for< 100M vectors, single-node> 100M vectors, distributed
Zvec is optimized for scenarios where your vector data fits comfortably in memory on a single machine (up to billions of vectors on modern servers). For larger-scale distributed deployments, consider client-server alternatives.

​
Architecture Overview

Zvec is built on Proxima, Alibaba’s production-grade vector search engine that powers large-scale AI applications. The architecture consists of:
  • Core engine (C++): High-performance vector indexing and search algorithms
  • Language bindings: Python and Node.js bindings via native extensions
  • Zero dependencies: Runs entirely in-process with no external services

​
Performance at Scale

Zvec delivers exceptional speed and efficiency, making it ideal for demanding production workloads. On a dataset of 10 million 768-dimensional vectors:
  • Query latency: < 10ms for 95th percentile
  • Indexing speed: Millions of vectors per second
  • Memory efficiency: Optimized data structures with optional quantization
For detailed benchmark methodology, configurations, and complete results, see our Benchmarks documentation.

​
Next Steps

Installation

Install Zvec for Python or Node.js

Quickstart

Build your first vector search application in 5 minutes

Build docs developers (and LLMs) love

Get started for freeTalk to us