Shy2593666979/AgentChat

AgentChat 是一个基于 LLM 的智能体交流平台，内置默认 Agent 并支持用户自定义 Agent。通过多轮对话和任务协作，Agent 可以理解并协助完成复杂任务。项目集成 LangChain、Function Call、MCP 协议、RAG、Memory、HITL、Skill、Milvus 和 ElasticSearch 等技术，实现高效的知识检索与工具调用，使用 FastAPI 构建高性能后端服务。

What it solves

AgentChat provides a complete, modern interface and backend for building and interacting with AI agents. It solves the problem of fragmented AI tools by integrating multi-model support, knowledge base retrieval (RAG), and tool calling into a single, deployable system with a professional UI.

How it works

The system uses a decoupled frontend (Vue 3) and backend (FastAPI) architecture. It leverages LangChain for orchestration and supports a three-layer memory architecture (short-term, historical summaries, and long-term preferences) to maintain context. It integrates the Model Context Protocol (MCP) for server extensions and uses vector databases like ChromaDB or Milvus for RAG capabilities.

Who it’s for

It is designed for developers and organizations looking to deploy a self-hosted, full-stack AI chat system that can handle complex tasks through agent collaboration and custom tool integration.

Highlights

  • Three-Layer Memory: Intelligent management of short-term context, historical summaries, and long-term user preferences.
  • Human-In-The-Loop (HITL): Allows manual intervention and confirmation at key decision nodes when generating MCP servers from OpenAPI specs.
  • Multi-Agent Collaboration: Supports sub-agents with reasoning and decision-making capabilities for complex task planning.
  • RAG Integration: Built-in knowledge base system with semantic chunking and vector retrieval.
  • Tool Ecosystem: Supports multi-turn tool calling (sequential dependency) and custom tool creation via Swagger/OpenAPI uploads.