Embed a vector database in your app to run local similarity search without a server
You can search billions of vectors in milliseconds without spinning up a server, anywhere your code runs. Zvec is a lightweight vector database that embeds directly into your application process, allowing you to perform similarity and full-text search simultaneously without complex infrastructure setup. The recent v0.8.0 release adds support for Intel Macs and Windows ARM64, and enables Chinese full-text search to work immediately without a separate dictionary.
What changed in v0.8.0
This update focuses on broadening platform support. You can now build and use the SDK on Intel-based Macs and Windows ARM64 environments, and the DiskANN index has been extended to Windows, Android, and iOS. Notably, the Jieba dictionary and HMM model are now built-in for Chinese full-text search, so you can use it immediately without separate dictionary files or path configuration. Additionally, peak memory usage has been reduced when creating IVF and DiskANN indexes, and HNSW RaBitQ configuration speed has improved through vector reuse and batch lookups.
How it searches vectors and text together
Zvec supports both dense and sparse vectors and offers various index types that scale from memory to disk. It natively supports full-text search on string fields using natural language or structured expressions, in addition to vector similarity search. You can combine these two with structured filters in a single query to perform hybrid search. Write-ahead logging ensures data is not lost during process crashes or power outages, and concurrent access is possible when multiple processes are read-only.
Installation and getting started
Official SDKs are available for multiple languages, including Python, Node.js, Go, Rust, Java, and Dart/Flutter. Python requires 64-bit Python 3.10 through 3.14, and Java requires Java 8 or higher. To start with Python, install it using the pip install zvec command, then define a collection schema, insert data, and run a similarity query. For searching code or documents, you can use zvec-grep (zg), a local-first CLI tool that integrates ripgrep, BM25, and vector search. If you want to explore data visually and debug queries without coding, you can use Zvec Studio.
In-process architecture that runs anywhere
Zvec operates as a library rather than a server, so it can be used anywhere your code runs, including laptops, servers, CLI tools, and even edge devices. It supports all major platforms, including Linux, macOS, Windows, Android, and iOS. This structure allows you to start searching locally and immediately without external network dependencies, with minimal setup required.
Things to check before using it
Zvec aims for production-grade performance validated within the Alibaba Group, but optimization for specific environments is still ongoing. Using Uniform quantization can avoid score reconstruction, but it may increase quantization error and reduce recall due to shared min/max ranges. Also, thread limits in index creation and parallel training jobs are applied more strictly, so you need to verify settings against your hardware specifications. The README does not explicitly mention license details or long-term support scope, so you should review the official documentation and release notes before applying it to commercial projects.
