Tsinghua built an AI classroom where AI agents teach and quiz you

What if you could upload any document — a research paper, a textbook chapter, even a Wikipedia article — and an AI classroom full of teachers and students appeared to teach it to you? That's exactly what OpenMAIC does.

Built by Tsinghua University's MAIC lab, OpenMAIC uses multiple AI agents (think of them as separate AI personalities) that work together to deliver a full learning experience: one agent lectures, another asks questions, a third draws on the whiteboard, and yet another grades your quizzes in real time.

One click. Full classroom.

You type a topic — "Teach me Python from scratch in 30 minutes" — or upload study materials. Within minutes, OpenMAIC builds a complete lesson with:

28 different ways it teaches

Under the hood, OpenMAIC has 28 different action types — speech, drawing shapes on the whiteboard, writing formulas, highlighting with a laser pointer, spotlight effects, and more. These are orchestrated by a system called LangGraph (a framework for coordinating multiple AI agents), which manages turn-taking and discussion flow between the AI agents.

The result feels less like reading a textbook and more like sitting in a classroom where the teacher knows exactly what you need to hear next.

From board games to stock analysis

The demo showcases surprising versatility:

Free to use, or run it yourself

You can try it right now at open.maic.chat — no installation needed. If you want full control, self-hosting takes three steps:

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

It works with OpenAI, Claude, Gemini, and DeepSeek. The team recommends Gemini 3 Flash for the best balance of speed and quality. One-click Vercel deployment and Docker are also available.

The academic backbone

This isn't a weekend hackathon project. The research behind OpenMAIC was published in the Journal of Computer Science and Technology (2026) under the title "From MOOC to MAIC: Reimagine Online Teaching and Learning through LLM-driven Agents." The project has earned 9,500 GitHub stars and is licensed under AGPL-3.0.

Who benefits most

Students who learn better from interactive experiences than static textbooks. Teachers who want to generate engaging lesson materials from their notes. Self-learners tackling unfamiliar topics — just describe what you want to learn, and an entire classroom materializes.

You can even export the generated lessons as PowerPoint files or interactive HTML — meaning teachers can use the AI to prepare materials, then present them in person.

Related Content — Get Started with Easy Claude Code | Free Learning Guides | More AI News