ragent
企业级 Agentic RAG 智能体 - 全链路覆盖文档解析、多路检索、意图识别、问题重写、会话记忆、MCP 工具调用与深度思考。面向真实业务场景,从 0 到 1 完整工程实现。
What is ragent?
企业级 Agentic RAG 智能体 - 全链路覆盖文档解析、多路检索、意图识别、问题重写、会话记忆、MCP 工具调用与深度思考。面向真实业务场景,从 0 到 1 完整工程实现。
Overview of ragent
What it is
Ragent is a production-grade Java AI application platform designed for Agentic RAG evolution. It covers the complete pipeline from document ingestion to intelligent question answering. The platform is described as an enterprise-level Agentic RAG agent with full-chain capabilities including document parsing, multi-path retrieval, intent recognition, question rewriting, conversation memory, MCP tool invocation, and deep reasoning.
Key features
- Hybrid retrieval: parallel recall via vector, keyword, knowledge graph, and web search, with deduplication, RRF fusion, and reranking.
- Question understanding: query term mapping, question rewriting and splitting, tree-structured intent recognition, and multi-knowledge-base routing.
- Model and tool support: model tiers, first-packet detection, circuit breaking and degradation, plus MCP tool discovery and invocation via AgentScope.
- Conversation memory: recent N turns combined with persistent summarization to control token costs while retaining key context.
- Traffic protection: Redis fair queuing and distributed concurrency control to prevent sudden request spikes from overwhelming model services.
- Knowledge closure: configurable ingestion pipeline, remote refresh, answer traceability, user feedback, traceability, and management backend.
Use cases
Enterprise knowledge Q&A, intelligent customer service, document processing, business system integration, multi-turn dialogue, and agent-based task execution that combines retrieval with tool calls.
Getting started
See the repository README for quick start instructions.
License and pricing
Licensed under Apache-2.0. Open source.
Based on the project's public README and repository metadata.