Qwen3 Retrieval

•
7
•

Provides a self-hosted server for Qwen3 embedding and reranking, featuring OpenAI-compatible APIs, a web console, and a Multi-Capability Protocol (MCP) server.

Qwen3 Retrieval offers a powerful self-hosted server for implementing advanced retrieval-augmented generation (RAG) workflows. It seamlessly integrates two independent vLLM 0.19.1 backends: one for efficient Qwen3-Embedding-4B vector recall and another for precise Qwen3-Reranker-0.6B document re-ranking. The server provides OpenAI-compatible APIs for easy integration into existing systems, alongside an interactive web console for visualizing embeddings with a 3D projector and analyzing results. With robust operational features like hot reloading, model caching, and detailed health checks, it empowers developers to manage and deploy sophisticated RAG pipelines efficiently.

Key Features

01Independent vLLM backends with hot reloading and health checks
025 GitHub stars
03OpenAI Compatible Embeddings API and vLLM Compatible Rerank API
04Integrated Web UI with Embedding analysis and 3D Vector Projector
05MCP Server for programmatic access to embedding, rerank, and projector tools
06Automatic model download and caching for Qwen3-Embedding-4B and Qwen3-Reranker-0.6B

Use Cases

01Integrating Qwen3 embedding and reranking capabilities into custom applications via standard APIs
02Analyzing and visualizing vector embeddings for data exploration and clustering
03Deploying self-hosted Qwen3 models for Retrieval Augmented Generation (RAG) applications