jrag

简体中文

English

Jrag is a Retrieval-Augmented Generation (RAG) and MCP tool integration platform based on Java Spring Boot. It aims to enhance the application capabilities of large language models in the Java ecosystem by combining retrieval, MCP tools, and generative AI model technologies. The platform supports access to various mainstream large language model interfaces, including Ollama and OpenAI, and integrates with Milvus vector database to provide efficient vector storage and retrieval services.

Contributors

One-Click Deployment with Docker

All Docker configurations are located in the docker/ directory. By default, it will start Milvus (v2.6.9) and Jrag.

1. Pull all dependency images (optional)

docker pull maven:3.8.8-amazoncorretto-21-debian
docker pull eclipse-temurin:21-jre
docker pull alpine/git
docker pull milvusdb/milvus:v2.6.9
docker pull debian:bookworm-slim

1. Pull frontend

shell

rm -rf jrag-starter/src/main/resources/dist
git clone -b dist https://github.com/jerryt92/jrag-ui.git jrag-starter/src/main/resources/dist

Windows

Remove-Item -Recurse -Force jrag-starter\src\main\resources\dist

git clone -b dist https://github.com/jerryt92/jrag-ui.git jrag-starter\src\main\resources\dist

1. Deploy

docker compose -f docker/docker-compose.yml up -d --build

Configurable options (docker/.env):

• JRAG_BASE_DIR: Host configuration/data root directory (default ~/jrag)
• COMPOSE_PROJECT_NAME: Container prefix (default jrag)
• UPDATE_UI: Whether to pull the latest dist from jrag-ui (true/false)
• JRAG_UI_REPO: UI repository address (default https://github.com/jerryt92/jrag-ui.git)
• JRAG_UI_BRANCH: UI branch (default dist)

Access:

• UI: http://localhost:30110/
• Health check: http://localhost:30110/v1/api/jrag/health-check

Host access within containers:

• macOS/Windows: host.docker.internal
• Linux: host.docker.internal (requires Docker 20.10+ and configuration of extra_hosts: ["host.docker.internal:host-gateway"])

Demo

Data Communication Encyclopedia Assistant

Data Communication Encyclopedia Assistant, based on Jrag, can answer various data communication-related questions.

Architecture

Demo

Purpose

So far, most open-source RAG platforms are implemented in Python. As a Java developer, I hope Jrag can better suit Java developers’ needs and provide more suitable LLM integration and applications for the Java ecosystem.

Features

• Multi-model support: Compatible with Ollama and OpenAI-style interfaces, allowing flexible switching between different large language models.
• Vector database integration: Supports Milvus vector database to meet performance requirements in various scenarios.
• Function Calling: Supports function calling, enabling LLMs to call APIs from other systems.
• MCP support: Supports MCP (Model Context Protocol) to standardize model tool invocation.
• MCP Client interacts with LLM using Function Calling technology instead of Prompt, saving token consumption.
• Java ecosystem optimization: Designed specifically for Java developers to simplify the integration and application of RAG technology in Java projects.
• JDK21: Jrag is developed based on JDK21, utilizing virtual threads to improve concurrent performance.
• Knowledge management: Provides knowledge base management functions, supporting operations such as adding, modifying, deleting, and hit testing of knowledge base content.

Interface

The interface style is dynamic, featuring frosted glass design and dark mode support.

Knowledge Management

To Be Improved

• Rerank: Provide reranking functionality to sort and filter retrieval results.
• Streamable HTTP transport layer adapted to MCP protocol (awaiting Spring AI release).
• Knowledge base maintenance: Provide knowledge base management functions, supporting operations such as creation, import, export, and deletion of knowledge bases.

Default Account Credentials

admin
jrag@2025

Frontend

jrag-ui

This site is open source. Improve this page.