Interactive classroom simulation with multi-AI agents

What you get

• Role-based multi-agent architecture: Dynamic interaction of LLM agents in the roles of teacher, questioning student, debater, and summarizer.
• Visual and audio classroom interface: Immersive pedagogical experience with virtual whiteboard, real-time Q&A stream, and text-to-speech (TTS).
• Customizable course curriculum: Instantly create interactive lessons by uploading your own PDF documents or text-based lecture notes.
• One-click simulation launch: Manage complex agent orchestration via a modern web interface without needing technical coding knowledge.
• Open-weight model compatibility: The freedom to connect any AI model you want via Ollama, vLLM, or cloud LLM providers.

Installation

git clone https://github.com/THU-MAIC/OpenMAIC.git
cd OpenMAIC
pnpm install

Running it

pnpm run dev
# Tarayıcıda http://localhost:3000 adresini açın

Technical architecture and working principle

• Chat Orchestration Engine: The central controller that manages which agent speaks when, the order of turns, and the context of the discussion.
• Memory and Context Management: Storing shared whiteboard content and student questions in short/long-term memory throughout the lesson.
• Real-Time Streaming via WebSocket: Delivering speech text, emotional expressions, and animations to the frontend interface without latency.

Multi-agent classroom dynamics and role simulations

• Socratic Debate Environments: Agents with diverse perspectives debate a topic to trigger the user's critical thinking.
• Personalized Tutor Support: Custom AI tutors that automatically adjust the difficulty level based on the user's learning pace.
• Research on Social Interaction Between Agents: Analyzing how large language models collaborate and share information in crowded group environments.

If you don't write code

Frequently asked questions

• Do I need a GPU to use OpenMAIC? If you are going to run your own local model (Ollama/vLLM), a GPU is recommended; however, it can be used directly with a standard computer via cloud APIs (OpenAI, Gemini, Groq).
• Can the user participate in the simulation via voice? Yes. Thanks to the WebRTC and voice recognition module, the user can join classroom discussions by speaking into their microphone.
• How many agents can be in a classroom at the same time? In the default configuration, an ideal interaction is achieved with between 3 and 8 agents; larger classes can be set up depending on system resources.
• In which formats can course content be uploaded? Plain text, Markdown, and PDF documents can be imported directly into the system's knowledge base.

Related dictionary terms

Links

• GitHub repository →
• Read in Turkish →