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、
What is 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。
Overview of AgentChat
What it is
AgentChat is a frontend-backend separated agent application platform. It connects model services and allows agents to be bound with tools, Skills, MCP, and knowledge bases. The platform includes a default agent and supports user-defined agents. It integrates LangChain, Function Call, MCP protocol, RAG, Memory, HITL, Skill, Milvus, and ElasticSearch.
Key features
- Flexible sessions: switch agents within the same session; asynchronously generate a 2–10 character topic title based on the first question.
- Agents: select models, configure prompts, and combine tools, MCP, Skills, and knowledge bases.
- Mars tasks: initiate tasks from the exploration page, view tool execution processes and Markdown reports.
- Tools and MCP: use built-in tools, import OpenAPI, or connect SSE / Streamable HTTP MCP services; support conversational MCP creation and human-in-the-loop confirmation.
- Skills: load six skill templates at startup: research report, data analysis, product documentation, code review, email writing, and knowledge organization.
- Knowledge bases: upload documents, vectorize them, bind to agents, and retrieve with reranking; full-text search can be enabled as needed.
- User resource isolation: each registered user gets independent copies of default agent, tools, Skills, and MCP; edits do not affect other users.
Use cases
- Initiate search, research, and report generation tasks from the exploration page, viewing the complete execution process and results.
- Use agents in conversations to assist with complex tasks by binding tools, MCP, Skills, and knowledge bases.
- Upload materials to knowledge bases for retrieval and reranking, enabling efficient information access.
- Manage model calls and token usage with a data dashboard, request tracing, and tool execution logs.
Getting started
bashgit clone --branch main https://github.com/Shy2593666979/AgentChat.gitcd AgentChat/src/backenduv sync --locked --python 3.12Create a local configuration file:bashcp agentchat/config.simple.yaml agentchat/config.local.simple.yamlEdit src/backend/agentchat/config.local.simple.yaml to set your model API key, base URL, and model name.Start the backend:bashexport AGENTCHAT_CONFIG=agentchat/config.local.simple.yamluv run --locked uvicorn agentchat.main:app --host 127.0.0.1 --port 7860In a separate terminal, start the frontend:bashcd src/frontendnpm cinpm run dev -- --host 127.0.0.1 --port 8090 --strictPort
License and pricing
This project is licensed under the MIT License. It is open source.
Based on the project's public README and repository metadata.