OpenMAIC (Open Multi-Agent Interactive Classroom) turns a topic or a set of documents into an interactive classroom: it generates slides, quizzes, interactive HTML simulations, and project-based learning (PBL) activities, and delivers them through AI teachers and AI classmates that speak, draw on a shared whiteboard, and hold discussions with the learner. It is a Next.js and React application with LangGraph orchestrating the agents, and it runs without a database by default, keeping course documents and learner state in browser storage.
Lesson generation is a two-stage pipeline: an outline stage analyses the input and produces a structured lesson outline, and a scenes stage turns each outline item into a slide, quiz, interactive module, or PBL activity. Input can be a text prompt or reference materials (PDF, Word, PowerPoint, spreadsheet, text, image, audio, and video) that configured extractors turn into content. Deep Interactive Mode adds five interactive UI types (3D visualization, simulation, game, mind map, and in-browser programming), and the AI teacher can operate these UIs to highlight areas, set conditions, and give hints.
Version 1.0.0 adds a Pro workbench: a chat-first workspace where an agent plans a curriculum, builds and revises every page, and works from uploaded materials or web search. Sessions are server-backed, survive restarts, and can be cancelled, resumed, and steered while running. The agent works through validated tools (plan curricula, patch a scene, reorder pages, extract materials, generate media and narration, import .pptx) and twenty built-in skills for lecture, workshop, vocational, and other teaching styles. The workbench is off by default and requires PostgreSQL.
Model and media providers are configured server-side. LLM providers include OpenAI, Azure OpenAI, Anthropic, Amazon Bedrock, Google Gemini, DeepSeek, Qwen, Kimi, MiniMax, Grok (xAI), OpenRouter, Doubao, Tencent Hunyuan, Xiaomi MiMo, GLM (Zhipu), Ollama, Lemonade (local LLM, image, TTS, and ASR), FunASR (local ASR), and any OpenAI-compatible API. Text-to-speech options include VoxCPM2 with voice cloning, and document parsing can route through MinerU. An OpenClaw skill generates and views classrooms from messaging apps such as Feishu, Slack, Discord, and Telegram, in hosted mode with an access code or against a self-hosted instance.
Features
- Classroom components: slides with voice narration, spotlight, and laser-pointer effects; quizzes with AI grading; HTML simulations; and PBL projects with roles, milestones, and deliverables
- Multi-agent interaction: classroom discussion, roundtable debate between agent personas, free Q&A, and a shared real-time whiteboard
- Deep Interactive Mode: 3D visualization, simulations, games, mind maps, and online programming, responsive across desktop, tablet, and mobile
- Agent workbench: durable sessions, session materials, course tools, and 20 built-in skills plus per-owner user skills
- Export: editable .pptx, self-contained interactive HTML, and classroom ZIP; exported classrooms inline external assets as data: URIs so they play offline
- Video export: an opt-in render-service container renders an MP4 with Chromium and FFmpeg
- Pluggable storage: browser storage by default; @openmaic/storage adds PostgreSQL-backed documents, learner runtime, assets, and agent sessions, with asset bytes in PostgreSQL or S3
- Provider controls: credentials stay server-side, ENABLED=false switches force off a capability, and unresolved model routes fail at startup
- Speech and search: multiple TTS and ASR providers, microphone input, and web search providers such as Brave, Baidu, Bocha, SearXNG, and MiniMax
- Deployment: Vercel, Docker Compose, and an ACCESS_CODE site password for shared deployments
- Localisation: 12 locales across 11 languages, plus dark mode
