Zvec — встраиваемая векторная база

★ 16K

Zvec is an open-source vector database that runs inside your application's process: no separate server and no configuration, built for semantic search, RAG and document retrieval. Its Alibaba authors say it has been battle-tested inside the company. Reach for it when you need vector search in a Python, Node.js, Go, Rust or Dart project without deploying a separate database, when vectors must stay local, when you are building RAG for an LLM agent, or when one query must combine vectors, full-text and structured filters. Inside are dense and sparse embeddings, HNSW, DiskANN and IVF-RaBitQ indexes, a PQ-INT8 quantizer, BM25 full-text search with an N-gram tokenizer, and write-ahead logging for durability. Several processes may read a collection at once, but writes are single-process exclusive. Install with pip install zvec (64-bit Python 3.10 to 3.14), npm install @zvec/zvec, cargo add zvec-rust, flutter pub add zvec, plus Go bindings. There is a zvec-grep CLI (ripgrep, BM25 and vector search) and the visual Zvec Studio. It runs on Linux (glibc and musl), macOS and Windows, with builds for Android and iOS. It is an embedded library, not a client-server system, and concurrent writes from several processes are not supported. Apache-2.0 license.