Project
RagentAI
Enterprise-grade Agentic RAG platform with MCP tools and full-stack observability.
StatusproductionRagentAI is an enterprise-level Agentic RAG intelligent Q&A platform covering the complete engineering workflow: knowledge ingestion, intent recognition, multi-channel retrieval, MCP tool invocation, streaming dialogue, and back-office operations.
Ingestion Pipeline
- Orchestrated document processing chain: Fetcher → Parser → Enhancer → Chunker → Indexer, with conditional skip and loop detection
- Multi-format support: PDF, DOCX, Markdown, HTML via Apache Tika 3.2
- Object storage: S3-compatible (RustFS) for document persistence
- Full lifecycle management: Knowledge base, document, and chunk CRUD with file upload and URL fetch
Retrieval & Intent
- Intent recognition: Tree-structured intent classification — KB Q&A, system interaction, MCP tool calling
- Multi-channel retrieval: Intent-directed retrieval + global vector retrieval + Rerank post-processing
- Hybrid search: Dense embeddings (Milvus 2.6) combined with sparse retrieval strategies
Generation & Conversation
- Streaming responses: SSE-based streaming with real-time answer generation
- Conversation memory: Summary compression + sliding history window
- Model routing: Multi-provider support with automatic fallback and health-based circuit breaking
Tech Stack
- Backend: Spring Boot 3.5 + MyBatis-Plus + Sa-Token + Redis/Redisson
- Frontend: React 18 + TypeScript + Vite 5 + Tailwind CSS + Radix UI + Zustand