Ragent
A production-oriented Java platform for building agentic retrieval-augmented generation applications, covering document ingestion, hybrid retrieval, intent understanding, session memory and MCP tool calling.
Ragent is a Java platform for building agentic RAG applications, covering the pipeline from document ingestion to question answering. It combines vector search, keyword search, knowledge-graph lookups and web search for retrieval, with deduplication, reciprocal rank fusion and reranking.
For understanding user queries it supports query rewriting and splitting, tree-based intent recognition and routing across multiple knowledge bases. It integrates with model providers and supports tool discovery and invocation through the Model Context Protocol via AgentScope, and keeps session memory by combining recent messages with a persisted summary to control token cost. It also includes Redis-based request queuing and concurrency control, an ingestion pipeline, remote refresh, answer traceability, user feedback, tracing and an administration console.
It is aimed at backend developers moving from traditional Java development into building production AI agent systems.