Shy2593666979/AgentChat

AgentChat – Modern AI‑Powered Conversational Platform

What it is – AgentChat is an open‑source, web‑based chat system built around large language models (LLMs). It provides a full‑stack UI (Vue 3) and a FastAPI backend that glue together LLM inference, retrieval‑augmented generation (RAG), tool calling, and multi‑agent coordination.

Key capabilities

• Smart agents – Sub‑agents can collaborate, reason, and make decisions; agents can be equipped with custom tools (via uploaded Swagger/OpenAPI specs) and with Skills that load progressive prompts.
• Three‑layer memory –
  1. Short‑term: keeps the most recent ~3000 tokens for immediate context.
  2. Historical summarisation: automatically condenses older dialogue.
  3. Long‑term: persists user preferences and habits for personalised conversations.
• RAG / knowledge base – Supports vector stores (ChromaDB, Milvus) and Elasticsearch for semantic search, enabling accurate document‑based Q&A.
• MCP server integration – Implements the Model‑Context‑Protocol, allowing external services to interact with the LLM context.
• Human‑in‑the‑loop (HITL) MCP generation – Users can iteratively refine an OpenAPI description through a conversational UI, with the system generating an MCP server configuration that can be manually approved at decision points.
• Tool ecosystem – 10+ built‑in utilities (e.g., weather lookup, image generation) and the ability to add custom tools.
• Admin & analytics – User authentication, role‑based permissions, and dashboards that show per‑agent usage, token consumption, and call counts.

Tech stack

• Backend: Python 3.12+, FastAPI, LangChain 1.0, MySQL, Redis, ChromaDB (or Milvus/Elasticsearch). Dependency management via Poetry/uv.
• Frontend: Vue 3.4+, Element Plus, Pinia, Vite 5, TypeScript.
• Deployment: Docker & Docker‑Compose scripts for one‑click production rollout; also supports local development.

Getting started

1. Clone the repo.
2. Edit docker/docker_config.yaml (database credentials, vector store choice, etc.).
3. Run docker-compose up --build -d for a full containerised stack, or start backend (uv sync && uv run uvicorn ...) and frontend (npm install && npm run dev) separately.
4. Open the web UI, log in, and begin creating agents, uploading knowledge bases, or linking custom tools.

Who might use it – Developers building internal AI assistants, SaaS teams needing a customizable chat UI, or researchers prototyping multi‑agent workflows and RAG pipelines.

License – MIT, so you can freely modify, redistribute, or embed AgentChat in commercial products.