OpenMAIC: An Open Source Platform for Generating Multi-Intelligent Body Interactive Teaching Classrooms with One Click

OpenMAIC (Open Multi-Agent Interactive Classroom) is an open source multi-intelligent body interactive classroom platform launched by the research team of Tsinghua University (THU-MAIC). It is developed based on the cutting-edge Large Language Model (LLM) and Multi-Agent Orchestration Architecture, and is able to automatically transform any user-input knowledge topic or uploaded document into a structured, visualized interactive digital classroom in one click. In the OpenMAIC-generated classroom, teachers and classmates played by AIs will take over the teaching process, and they can not only read the courseware aloud by voice, but also hand-draw diagrams and write mathematical formulas in real time on the shared whiteboard, as well as engage in real-time Q&A and roundtable debates with human learners.

In order to break the limitations of passive listening in traditional online education, OpenMAIC has introduced the “Deep Interactive Mode”. This mode supports the rendering of 3D visualization models directly in the browser, the provision of physical process simulation with dynamically adjusted parameters, the built-in code editor for online programming verification, and the automatic generation of mind maps related to the knowledge framework. In addition, the platform supports seamless interface with OpenClaw intelligent plug-ins, allowing users to command the AI assistant to generate teaching links directly in daily chat software such as Flybook, Slack, Telegram, and so on. The platform provides the ultimate data flexibility. Whether docking international mainstream AI models or conducting completely offline local model deployment, OpenMAIC can efficiently meet the teaching needs of diversified scenarios, such as fragmented self-study, professional thesis study and corporate training.

Function List

• Generate structured courses with one click: Enter a learning topic or upload a reference document (PDF support), and the back-end algorithm automatically performs two-stage processing to generate a complete course outline, graphic slides, and post-course quiz content in minutes.
• Multi-Intelligence Body Classroom Interaction SystemAI teacher and AI classmates with different settings are built-in. It supports realistic interaction scenarios such as automatic explanation, initiating multi-character roundtable debates, real-time classroom question and answer, and free exploration.
• Deep Interactive Scene Rendering Mode：
  • 3D Visualization: Automatic rendering of 3D view models for abstract structures, with support for mouse drag and drop for all-around viewing.
  • Dynamic Simulation: Automatically generate experimental environments with physical or process mechanisms, with adjustable parameters to observe real-time changes.
  • Interactive Learning Games: Built-in mini knowledge consolidation levels to deepen memorization points in interactive challenges.
  • Online Programming Lab: Provide a compilation environment for writing, modifying, and running code on the fly on the web side.
  • Automated mind maps: Instantly generates a scalable map of the system's knowledge structure based on the knowledge logic.
• Intelligent Dynamic Whiteboard with Speech Synthesis: AI is able to explain knowledge in a human voice with built-in TTS technology and use indicators on the whiteboard to highlight key points, step-by-step hand-drawn flowcharts and act out complex formulas.
• Multi-format asset export: Supports one-click export of the generated full course to a secondary editable PowerPoint (.pptx) file that retains layout and formatting, one-page offline HTML with interactions, or a ZIP package of classroom data.
• Highly compatible open LLM accessIt supports domestic and international mainstream APIs such as OpenAI, Google Gemini, Anthropic, DeepSeek, Smart Spectrum GLM, MiniMax, etc., and supports connecting local private models through Ollama.
• Chat software shortcut integration (OpenClaw): Seamless access to instant messengers, allowing bots to automatically queue up to generate classroom links in the background by sending chat commands in Flybook, Slack, and Discord.

Using Help

OpenMAIC provides a high degree of flexibility, whether you are a user who expects to quickly experience the generation ability, or a developer who needs to carry out the deployment of private data isolation, you can find the corresponding path of use. Below you will find a comprehensive guide to installation, configuration and core functionality. Follow the steps and start your journey of intelligent and interactive learning.

🌟 I. Quick Installation and Localization Deployment Guide

In order to obtain the most complete deep interaction capability without delay, it is strongly recommended that users with some development foundation use a local environment for deployment.

1. Preparation of the base environment
Before cloning the project, make sure that the following base software is properly installed and configured in your operating system environment:

• Node.js: The versioning requirements must >= 20.9.0, providing operational support for both back-end and front-end components.
• pnpm: powerful front-end package management tool, version requirements >= 10。

2. Obtaining source code and installing dependencies
Open your terminal (Terminal / PowerShell / iTerm2) and execute the following commands in sequence to clone the official project source code locally and automate the installation of all required dependent environment packages:

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

3. Configuring Environment Variables and the Large Language Model (LLM)
The core of multi-intelligence interaction relies on the powerful logic capability of large language models, and OpenMAIC supports rich model interfaces.
First, copy the generated environment variables file in the root directory:

cp .env.example .env.local

Open with a text editor .env.local file, fill in the Big Model API key you are using.
Examples of common large model configurations:

• OpenAI：OPENAI_API_KEY=sk-...
• Smart Spectrum GLM：GLM_API_KEY=您的密钥 and configured underneath GLM_BASE_URL=https://open.bigmodel.cn/api/paas/v4
• local disconnection model: If you have extremely high data privacy needs, you can also bind in the environment file the locally running Ollama Service port address.

💡 Best Experience Recommendations: To obtain an optimal balance between the quality of content generation and the speed of reasoning, it is recommended to use the Google Gemini 3 Flash. If you need to set this model as the global default, you can add a configuration file to the DEFAULT_MODEL=google:gemini-3-flash-preview。

4. Launching of Platform services
Run the development environment startup command in the terminal:

pnpm dev

At this point the system will start the local server in the background. Open a modern browser (Chrome or Edge recommended) and visit the address http://localhost:3000If you want to see the main interface of OpenMAIC, you can see the main interface of OpenMAIC.

📌 Supplemented by other deployment modalities：

• Docker One-Click Deployment: Modified .env.local After the key, the terminal executes the docker compose up --build It is possible to achieve stable containerized operation with environmental isolation.
• Setting up access encryption protection: If you are deployed in a public access environment, the .env.local become a member ACCESS_CODE=您的专属安全密码. This way all external visitors are blocked to ask for authentication, preventing theft of interface credits.

🎓 II. Quickly generate and experience an interactive classroom from zero base

Once configured and successfully entered into the web interface, you will be presented with the minimalist Course Workbench. Please follow the three standard processes below:

Step 1: Enter a core topic or upload a knowledge document
In the large text input area on the home page, explicitly type in what you want to learn. For example: “Please explain to me the principle of supply and demand in macroeconomics” or “How to develop a crawler script using Python”.
If you have a specialized textbook or professional paper, you can upload a PDF document by clicking the “Attachments” button in the lower left corner, and the system will use the built-in algorithms to identify and extract text, tables and complex mathematical formulas from the document.

Step 2: Review and fine-tune the generated course outline
Click on the “Generate” button and the AI will output a copy of theStructured Course Outline (Outline)。
The outline is presented as a tree node, where you can directly manipulate it as if you were editing a catalog:

• Click on the “+” sign to add a new Knowledge Extension subsection.
• Click on the trash icon to delete unwanted topics of discussion.
• Double-click on a node to modify the direction of exploration of specific chapter content.

Step 3: Content Rendering and Entering the Classroom
After confirming the outline, the system automatically entersScene Rendering (Scenes) Phase.. Multiple intelligent bodies work in parallel to generate voice-reading scripts, slide layout illustrations, and quiz questions, respectively.
Wait for the progress bar to fill up and click “Start Class” on the interface. At this point you will enter a digital virtual classroom.

• Immersive listeningThe main visual of the screen shows slides with highlighted animated laser pointer instructions, and the AI teacher automatically uses TTS technology to explain the key points page by page in a human voice.
• Interruption of the questioning mechanismAI Teacher: When you hear a concept that's not easy to understand, type a question into the chat box below (or use the microphone to voice it) and the AI Teacher will pause the slide show to answer your question directly and in layman's terms. If you need a demonstration, it will also actively cut out theDynamic Whiteboard (Whiteboard)The program is designed for you to draw data flow charts or derive complex formulas in handwriting.

🎮 III. Deep Interactive Mode (DIM) exploration

When confronted with subjects that require hands-on practice, such as programming, physics, or three-dimensional structures, OpenMAIC automatically brings up the deep interactive UI in the corresponding section:

• When learning about physical processes: it will pop upPhysics Simulator. You can slide the numerical levers in the page (e.g., to change the acceleration of gravity, temperature) and observe the system's experimental dynamics in real time for feedback.
• When learning computer programming: The screen will be automatically cut out in a split screenOnline Code Editor (Online Programming)The system initializes a base code for you. The system initializes a basic code for you, you can rewrite it directly on the web page and click “Run” to see the output of the terminal, and the AI will also guide you how to fix the code in real time when it reports an error.
• When studying the human body or molecular structure: By3D structure visualizationcontrol, you can drag and drop with the left mouse button for a 360° view of the profile.
• When it comes to chapter review summaries: Trigger **Global Mind Map** mode, clearly displaying the knowledge mapping network of this section, and clicking through the nodes to review the past teaching content.

📤 IV. One-click sharing and exporting of multi-format classroom files

At the end of a great interactive session, it is critical to assetize and share. Click on the top right corner of the page to Export button to bring up the export options:

1. Export as PowerPoint (.pptx)The system converts each page of the generated slides into a native PPT with the formatting fully preserved, the generated graphical elements and images are not converted to dead images, and the formulas within the course are accurately converted into the native Office Math formula format, which is extremely convenient for secondary modifications later on in the course for school defenses or conference presentations.
2. Export to Interactive HTMLIf you want to keep the simulations, interactive quizzes, etc., you can export this stand-alone HTML one-page file. It works perfectly in your local browser without the need for the Internet, making it the ideal offline review library.
3. Export as Classroom ZIP: The system comes packaged with a full source package including course data structures, media voice material, and is suitable for sharing with friends for importing into their OpenMAIC platforms for reproduction in its original form.

By following the help guide above, you will be able to master the system logic of OpenMAIC without any obstacles and fully unleash the amazing efficiency of multi-intelligence education!

application scenario

1. Individual zero-based self-study and immersion exploration
   Whether it's learning a new programming language or understanding complex scientific principles, users simply enter their learning objectives and the platform generates a structured curriculum with graphical explanations, code walkthrough environments, and knowledge quizzes.The AI tutor patiently explains the lessons by voice, and assists the user in learning and practicing through the built-in programming environments.
2. In-depth reading of specialized foreign language literature and analysis of financial reports
   Researchers or financial analysts can directly upload hundreds of pages of English-language full-length papers or financial report PDF documents. The multi-intelligence system can quickly extract the core data logic, generate multi-perspective discussion classes, and dynamically draw charts to demonstrate data relationships through the whiteboard, significantly reducing the threshold of understanding complex information.
3. Enterprise standards business training and interactive guidance
   Enterprise managers can turn long and boring employee onboarding manuals or software operation SOP documents into interactive AI online classrooms with a single click. New employees can ask questions to the AI tutor at any time during the training process and verify learning outcomes through the platform's built-in accompanying tests, enhancing the fun and knowledge retention rate of the training.
4. K-12 science visualization science and virtual experiment demonstration
   Educators can use this tool to transform abstract and complex concepts from textbooks into vivid and three-dimensional immersive laboratory lessons. Using the “3D Visualization” and “Physics Simulator” in the Deep Interaction Mode, students can adjust the parameters and observe the structural changes on the screen, which fully stimulates their interest in exploring science knowledge.

QA

1. What large language models (LLMs) does OpenMAIC support?
   OpenMAIC has strong interface compatibility, and currently supports calling most of the mainstream big models on the market by configuring API keys, including OpenAI (GPT series), Anthropic (Claude), Google Gemini, DeepSeek, Smart Spectrum GLM, MiniMax, Grok, and ByteHopper Doubao etc. It also supports access to local services such as Ollama to run completely offline.
2. If organizations want to protect local data privacy, how can they achieve a complete absence of running on the cloud?
   The platform has the ability to run purely locally. You can start by running the platform locally on your device via the Ollama Deploy open source large language models (e.g. Llama 3, etc.), and then deploy OpenMAIC backbone code locally or privatize Docker within the LAN. Under this architecture, the entire process of course generation analysis (including document analysis and extraction) is done in a closed loop within the LAN, and there is absolutely no data outflow.
3. Are the PPT files exported through the platform static images that can be subsequently modified?
   The exported PowerPoint (.pptx) files are fully editable, high-quality documents, and OpenMAIC's export engine preserves the native text boxes, shapes, and inserted image resources. Not only that, the mathematical calculations and derivations generated by the system can be converted into native LaTeX or Office Math format, so that you can modify, add or delete them as you like in the local software on your computer.
4. Can I use it out of the box if I don't have the technical base to deploy the code?
   OpenMAIC integrates natively with the OpenClaw Tools. If you don't know how to deploy it, you can use the Hosted Mode, which requires you to get a one-time Access Code from open.maic.chat, and then send a natural language command to the AI bot in popular chat programs such as Flybook, Slack, Telegram, or Discord, which will queue up a course in the cloud and return a web link directly to you where you can open the online class. The system will queue up the course in the cloud and return a link to the webpage where you can open the online class.
5. Does the system support automatic grading and correcting of quiz questions generated in the classroom?
   Supports automatic grading of assessments. When the lesson proceeds to the quiz section, the AI teacher will automatically present single-choice, multiple-choice or short-answer questions on the screen. After you click on the options and submit your answers online, the multi-intelligent body will evaluate your choices in the background, not only scoring in real time, but also giving you targeted, spoken principles for correcting the wrong part of the analysis.