What is this?
Alibaba's open-source, in-process vector database — a lightweight, lightning-fast similarity-search engine you embed directly into your application, battle-tested inside Alibaba Group.
Alibaba's open-source, in-process vector database — a lightweight, lightning-fast similarity-search engine you embed directly into your application, battle-tested inside Alibaba Group.So is this a server I deploy, or something that lives inside my app?A library, not a server: it runs in-process, pure local with no config, wherever your code runs — notebooks, servers, CLI tools, even edge devices.
A library, not a server: it runs in-process, pure local with no config, wherever your code runs — notebooks, servers, CLI tools, even edge devices.How do I add it to a project?Install an official SDK: pip install zvec (64-bit Python 3.10-3.14), npm install @zvec/zvec, cargo add zvec-rust, Go bindings, or flutter pub add zvec. Zvec Studio also lets you browse data and debug queries without code.
Install an official SDK: pip install zvec (64-bit Python 3.10-3.14), npm install @zvec/zvec, cargo add zvec-rust, Go bindings, or flutter pub add zvec. Zvec Studio also lets you browse data and debug queries without code.Do I need to install anything else first?Just the runtime for your SDK of choice — for example, 64-bit Python 3.10-3.14 for the pip package. No separate database server, service, or config files are required.
Just the runtime for your SDK of choice — for example, 64-bit Python 3.10-3.14 for the pip package. No separate database server, service, or config files are required.Can several of my services write to one collection at the same time?No — writes are single-process exclusive: many processes can read a collection simultaneously, but only one can write. Data safety itself is covered, since WAL guarantees persistence across crashes and power failures.
No — writes are single-process exclusive: many processes can read a collection simultaneously, but only one can write. Data safety itself is covered, since WAL guarantees persistence across crashes and power failures.Which platforms can it run on?Linux (x86_64 and ARM64, glibc and musl/Alpine), macOS (ARM64 and x86_64), and Windows (x86_64); prebuilt SDK binaries for Android and iOS ship with every release.
Linux (x86_64 and ARM64, glibc and musl/Alpine), macOS (ARM64 and x86_64), and Windows (x86_64); prebuilt SDK binaries for Android and iOS ship with every release.Does my data ever leave my machine?Nothing leaves your machine. Zvec is pure local with no servers — all indexing and search happens inside your own process.
Nothing leaves your machine. Zvec is pure local with no servers — all indexing and search happens inside your own process.What kinds of search does it actually support?Dense and sparse embeddings with multi-vector queries, native keyword-based full-text search, and hybrid search that fuses vector similarity, FTS, and structured filters in a single query. Vector index types scale from memory to disk.
Dense and sparse embeddings with multi-vector queries, native keyword-based full-text search, and hybrid search that fuses vector similarity, FTS, and structured filters in a single query. Vector index types scale from memory to disk.How fast is it, and how much data can it handle?It searches billions of vectors in milliseconds. v0.7.0 productionizes DiskANN, adding Linux and macOS ARM64 support plus an iouring async I/O backend that automatically falls back to the best available option.
It searches billions of vectors in milliseconds. v0.7.0 productionizes DiskANN, adding Linux and macOS ARM64 support plus an iouring async I/O backend that automatically falls back to the best available option.What is zvec-grep (zg)?A local-first workspace search CLI that unifies ripgrep, BM25, and vector search behind one command, built for humans and AI agents. Zvec also serves as a file store backend for ReMe, a memory management kit for agents.
A local-first workspace search CLI that unifies ripgrep, BM25, and vector search behind one command, built for humans and AI agents. Zvec also serves as a file store backend for ReMe, a memory management kit for agents.Which topics is it in?hnsw, faiss, similarity-search, llm-memory, local, hybrid-search, embedded-database, search-engine, flutter, vector-database, semantic-search, embeddings
hnsw, faiss, similarity-search, llm-memory, local, hybrid-search, embedded-database, search-engine, flutter, vector-database, semantic-search, embeddings