Zvec
Zvec is Alibaba's open-source, in-process vector database that embeds directly in your application for low-latency similarity search in RAG and agent memory, with HNSW, IVF and DiskANN indexes, quantisation, full-text search and builds for Linux, macOS, Windows, Android and iOS.
Updated
✨ Try AgentScout's free AI image generator →Good for: research, data analysis, working with documents, memory for agents, working with databases
Use cases
About Zvec
Zvec is listed in our memory & RAG category and is aimed at builders. Zvec is Alibaba's open-source, in-process vector database that embeds directly in your application for low-latency similarity search in RAG and agent memory, with HNSW, IVF and DiskANN indexes, quantisation, full-text search and builds for Linux, macOS, Windows, Android and iOS. Pricing, as the maker states it: Open source. Maturity: Mature. We last checked the official link on 2026-11-09.
What is Zvec?
Zvec is Alibaba's Apache-2.0 vector database that runs inside your process. The README says there is no separate server to deploy. It stores dense and sparse vectors, supports multi-vector queries, and names index types that cover memory and disk, including HNSW, IVF and DiskANN. It also has full-text search and a hybrid query that combines vector similarity, keywords and structured filters. A write-ahead log is how it keeps data across a crash. Several processes may read one collection. Writes are single-process.
Official SDKs in the README are Python (pip install zvec, 64-bit Python 3.10 through 3.14), Node.js (@zvec/zvec), Go, Rust, Java 8 or newer, and Dart or Flutter. Platforms listed are Linux, macOS, Windows, Android and iOS. The v0.8.0 notes add Intel Mac and Windows ARM64 builds and extend DiskANN to Windows, Android and iOS. Related tools, zvec-grep and Zvec Studio, are separate repositories. The README's line about searching billions of vectors in milliseconds is the project's description, not a benchmark AgentScout re-ran.
Sources: github.com/alibaba/zvec, github.com/alibaba/zvec/blob/main/README.md, zvec.org/en/docs/db/quickstart and our directory listing, checked 2026-11-09.
Who is Zvec for?
Developers embedding similarity search in an application, a notebook, a CLI or an on-device agent, who want the index in-process instead of a vector service to operate.
Pros and cons
Pros
- Install and call it from the application. The README's Python quick start creates a collection, inserts documents and searches.
- Full-text and vector conditions can be one query.
- The licence is Apache-2.0, including the SDKs the README points at.
Cons
- Writes are exclusive to one process. Concurrent writers are not the documented model.
- Uniform quantization can cut recall. The v0.8.0 notes say a shared min/max range can raise quantization error.
- Chinese full-text search in v0.8.0 embeds a Jieba dictionary. You do not configure that path, but tokenization still follows that dictionary.
- zvec-grep and Zvec Studio are not part of this repository.
Setup
- Install a SDK. For Python, use 64-bit Python 3.10 through 3.14 and pip install zvec. Node is npm install @zvec/zvec. Java needs Java 8 or newer.
- Follow the quickstart at zvec.org to define a collection, insert vectors and search.
- Pick an index from the vector-index docs. DiskANN is the disk-scale option the README names.
- If you need keyword search over a repo, look at zvec-grep separately. It is not installed with the database library.
Zvec at a glance
- Open source: Yes (Apache License 2.0)
- GitHub: alibaba/zvec · 16,219 stars (as of Nov 9, 2026)
Key features
- In-process library. No server process in the default install
- Dense and sparse vectors, HNSW, IVF, DiskANN, full-text search and hybrid queries
- Write-ahead log, multi-process readers, single-process writer
- Python, Node.js, Go, Rust, Java and Dart or Flutter SDKs
- Apache-2.0. Linux, macOS, Windows, Android and iOS builds
Zvec alternatives
Similar memory & RAG tools: MarkItDown, Claude-Mem, RAGFlow, Headroom, Mem0. Browse all memory & RAG tools →
Zvec FAQ
What is Zvec?
Zvec is Alibaba's Apache-2.0 vector database that runs inside your process. The README says there is no separate server to deploy. It stores dense and sparse vectors, supports multi-vector queries, and names index types that cover memory and disk, including HNSW, IVF and DiskANN. It also has full-text search and a hybrid query that combines vector similarity, keywords and structured filters. A write-ahead log is how it keeps data across a crash. Several processes may read one collection. Writes are single-process.
Who is Zvec for?
Developers embedding similarity search in an application, a notebook, a CLI or an on-device agent, who want the index in-process instead of a vector service to operate.
Is Zvec free?
Zvec's pricing, as the maker states it: Open source. Open-source tools are free to use, though you may still pay for model API usage or a hosted plan. Check the official site for current plans.
Is Zvec open source?
Yes. Zvec is released under the Apache License 2.0, according to its GitHub repository.
What are the pros and cons of Zvec?
Pros: Install and call it from the application. The README's Python quick start creates a collection, inserts documents and searches., Full-text and vector conditions can be one query., The licence is Apache-2.0, including the SDKs the README points at. Cons: Writes are exclusive to one process. Concurrent writers are not the documented model., Uniform quantization can cut recall. The v0.8.0 notes say a shared min/max range can raise quantization error., Chinese full-text search in v0.8.0 embeds a Jieba dictionary. You do not configure that path, but tokenization still follows that dictionary., zvec-grep and Zvec Studio are not part of this repository.
How do I set up Zvec?
1. Install a SDK. For Python, use 64-bit Python 3.10 through 3.14 and pip install zvec. Node is npm install @zvec/zvec. Java needs Java 8 or newer. 2. Follow the quickstart at zvec.org to define a collection, insert vectors and search. 3. Pick an index from the vector-index docs. DiskANN is the disk-scale option the README names. 4. If you need keyword search over a repo, look at zvec-grep separately. It is not installed with the database library.
What are alternatives to Zvec?
Similar tools include Chroma, LanceDB, Activeloop, xerj, txtai and Oracle AI Developer Hub.
Zvec ratings and reviews
Be the first to rate Zvec: tap 1 to 5 stars below. A written review is optional.
Zvec alternatives
Other memory & RAG tools in our directory. See all memory & RAG tools →
Chroma
Open-source search and vector database widely used for agent memory and RAG.
LanceDB
Open-source multimodal vector database for search, RAG and agent memory.
Activeloop
Deep Lake, a multimodal database for AI that stores vectors, images, text and video so agents and RAG apps can search and retrieve them.
xerj
A local-first search engine and MCP server for agents: one Rust binary indexes code, docs, logs, CSV/JSONL and PDFs for full-text, vector and hybrid search plus agent memory, behind an Elasticsearch-compatible API.
txtai
All-in-one Python framework for semantic search, LLM orchestration and agents built on an embeddings database of vector, graph and relational indexes.
Oracle AI Developer Hub
Technical resources for AI developers to build applications, agents, and systems using Oracle AI Database and OCI services
Get your badge
Built Zvec? Optional. Showing this badge does not change your rank or earn a Featured spot. It links to this page. About badges