Repolode
Submit
Shy2593666979

Shy2593666979/AgentChat

A self-hosted agent platform for chatting with customizable LLM agents that bind tools, MCP servers, Skills and knowledge bases.

View on GitHubWebsite
Stars
931
Forks
114
Open issues
30
Last commit
21 days ago

What is it?

AgentChat is a front-end and back-end separated platform for building and running LLM agents. You connect a model service, then configure each agent with a prompt, built-in or OpenAPI-imported tools, MCP servers over SSE or Streamable HTTP, Skill templates, and knowledge bases that are vectorized and retrieved with reranking.

The platform is designed to run locally with a lightweight setup: SQLite, in-memory cache, local files and Chroma by default, with MySQL, Redis, MinIO and Milvus available through configuration. Users can switch agents within one conversation, launch longer search-and-report tasks from an exploration page, and review tool execution steps and Markdown reports. Each registered user receives independent copies of default agents, tools, Skills and MCP entries.

The project also records model and token usage, offers a dashboard, request tracing and tool timing logs, and streams messages and run states through an AG-UI adapter. It is built on LangChain and FastAPI, with a Vue 3 frontend and an MIT license.

Key features

  • 01Chat sessions that let you switch agents mid-conversation and auto-generate topic titles
  • 02Agents configurable with prompts, tools, MCP servers, Skills and knowledge bases
  • 03Knowledge base upload, vectorization, retrieval and reranking, with optional Elasticsearch full-text search
  • 04MCP server connections over SSE or Streamable HTTP, plus OpenAPI tool import
  • 05Token usage statistics, dashboard, request tracing and tool timing logs

Use cases

  • Building a private assistant with custom agents backed by your own model API key
  • Connecting an internal knowledge base to an agent for document Q&A
  • Experimenting with MCP servers and tool calling in one web interface
  • Running multi-step research tasks that produce a Markdown report

Quick start

  1. 1

    Install uv, Python 3.12 and Node.js 22 as listed in the README

  2. 2

    Clone the repo and install backend dependencies

    git clone --branch main https://github.com/Shy2593666979/AgentChat.git && cd AgentChat/src/backend && uv sync --locked --python 3.12
  3. 3

    Copy the simple config and set your chat model API key in it

    cp agentchat/config.simple.yaml agentchat/config.local.simple.yaml
  4. 4

    Start the backend server

    export AGENTCHAT_CONFIG=agentchat/config.local.simple.yaml && uv run --locked uvicorn agentchat.main:app --host 127.0.0.1 --port 7860

Things to know

  • Knowledge bases need a separately configured embedding and rerank model; the chat model key is not enough.
  • Some external tools require their own API keys.
  • Self-hosting requires setting up a Python backend and a Node.js frontend in separate terminals.
  • The Docker setup has not been verified by the maintainers for this release.

In collections

MCP servers

Topics

#agent#llm#mcp#rag#langchain#fastapi#vue#knowledge-base#ag-ui#agent-memory

FAQ

What is AgentChat?
A self-hosted agent platform for chatting with customizable LLM agents that bind tools, MCP servers, Skills and knowledge bases. AgentChat is a front-end and back-end separated platform for building and running LLM agents. You connect a model service, then configure each agent with a prompt, built-in or OpenAPI-imported tools, MCP servers over SSE or Streamable HTTP, Skill templates, and knowledge bases that are vectorized and retrieved with reranking.
Is AgentChat free to use? What license does it have?
AgentChat is open source under the MIT license. Check the license terms before using it in commercial products.
Is AgentChat still maintained?
Actively maintained. The last commit to AgentChat was on Oct 17, 2026.
What can I use AgentChat for?
Building a private assistant with custom agents backed by your own model API key Connecting an internal knowledge base to an agent for document Q&A Experimenting with MCP servers and tool calling in one web interface Running multi-step research tasks that produce a Markdown report
How do I install AgentChat?
Install uv, Python 3.12 and Node.js 22 as listed in the README Clone the repo and install backend dependencies: git clone --branch main https://github.com/Shy2593666979/AgentChat.git && cd AgentChat/src/backend && uv sync --locked --python 3.12 Copy the simple config and set your chat model API key in it: cp agentchat/config.simple.yaml agentchat/config.local.simple.yaml Start the backend server: export AGENTCHAT_CONFIG=agentchat/config.local.simple.yaml && uv run --locked uvicorn agentchat.main:app --host 127.0.0.1 --port 7860
Where can I get AgentChat?
The source code and installation instructions are on GitHub: https://github.com/Shy2593666979/AgentChat

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects