Agenten & Multi-Agenten

Zvec: An Embedded, In-Process Vector Database

alibaba/zvec

Zvec is an Apache-2.0-licensed vector database designed to run directly inside applications. It supports dense and sparse vectors, full-text and hybrid search, structured filtering, local durable storage, and SDKs for several programming languages.

★ 15,4KSterne
⑂ 973Forks
76Offene Issues
C++Sprache
Apache-2.0Lizenz
Q@project.QualityScoreRedaktionelle Bewertung

Übersicht

Developed under the Alibaba GitHub organization, Zvec targets low-latency similarity search without a separate database server. Its primary implementation language is C++. The repository describes support for memory-to-disk vector indexes, multi-vector queries, full-text search, hybrid retrieval, write-ahead logging, and concurrent readers. The supplied metadata reports 15,360 stars, 973 forks, 76 open issues, and an active, non-archived main branch.

Wichtige Funktionen

  • In-process operation without a separate database server
  • Dense and sparse vector support
  • Multi-vector queries and multiple vector index types
  • Native full-text search
  • Hybrid vector, full-text, and structured-filter queries
  • Write-ahead logging for durable local storage
  • Multiple concurrent readers with single-process-exclusive writes
  • Group-by top-K search across Flat, HNSW, HNSW-RaBitQ, and sparse indexes
  • INT8 and INT4 quantization with optional random rotation
  • Unicode-aware tokenization, character folding, and Snowball stemming for 34+ languages
  • Officially listed SDKs or bindings for Python, Node.js, Go, Rust, and Dart/Flutter

Voraussetzungen, Installation und Schnellstart

Python: `pip install zvec`. Node.js: `npm install @zvec/zvec`. Rust: `cargo add zvec-rust`. Dart/Flutter: `flutter pub add zvec`. Go bindings are available at https://github.com/zvec-ai/zvec-go, but an installation command is not stated. Source-build instructions are available at https://zvec.org/en/docs/db/build/.

Nutzung

Python quick start: import `zvec`; create a `CollectionSchema` with a `VectorSchema`; call `zvec.create_and_open` with a local path; insert `zvec.Doc` records; then call `collection.query` with a `zvec.Query`, a query vector, and `topk`. Results are described as relevance-sorted records containing an ID and score. See https://zvec.org/en/docs/db/quickstart/ for the documented walkthrough.

Modellkompatibilität und Anwendungsfälle

Zvec stores and searches embeddings but the supplied repository material does not name compatible embedding models, LLM providers, or MCP integrations. Its RAG, agent-memory, and LLM-memory relevance comes from local dense, sparse, full-text, and hybrid retrieval rather than a stated dependency on a particular model.

Lizenz- und Risikohinweise

The repository metadata identifies the license as Apache-2.0. License reference: https://api.github.com/licenses/apache-2.0. Users should review the repository license and dependency terms before redistribution.

Veröffentlichung und Wartung

The README highlights v0.6.0, dated July 20, 2026. It adds group-by search, optional random-rotation quantization for INT8/INT4, an enhanced full-text pipeline, block-max skipping for conjunction queries, a DiskANN C API, and stability fixes. Release details: https://github.com/alibaba/zvec/releases/tag/v0.6.0. The supplied repository metadata records the latest push as August 3, 2026.

Firecrawl

firecrawl/firecrawl

★ 161,1KTypeScript

LangChain

langchain-ai/langchain

★ 143,6KPython

RAGFlow

infiniflow/ragflow

★ 86,7KGo