Most online courses are still passive; OpenMAIC flips that assumption by orchestrating multiple LLM-driven agents to generate and run interactive lessons on demand. The project combines a two-stage generation pipeline with a LangGraph-based multi-agent runtime so a single prompt or uploaded document becomes a playable classroom with narrated slides, quizzes, simulations and PBL workflows.
What Sets It Apart
- Two-stage pipeline + orchestration: input → structured outline → per-scene content. This separation lets the system first plan a lesson and then have specialized agents produce slides, quizzes, whiteboard steps and interactive HTML scenes.
- Multi-agent, real-time playback: teacher, TAs and classmates are agent personas that can speak (TTS), draw on a shared whiteboard, run discussions, and take turns—enabling roleplay, debates and guided exploration instead of a single monologue.
- Export & offline play: generated classrooms can be exported to editable .pptx, self-contained interactive HTML or a classroom ZIP that inlines assets for air-gapped playback—useful for education deployments with limited internet access.
- Flexible provider support and local-first options: supports many LLM/TTS/ASR providers and offers a Lemonade local provider and VoxCPM adapters for on-prem or offline speech synthesis.
Who It's For
Great fit if you are an educator, edtech product team, or researcher who wants to prototype or deploy active, social learning experiences from minimal inputs (a topic or a PDF). It scales from one-click demo classrooms to richer Deep Interactive Mode with 3D visuals, simulations and in-browser coding. Look elsewhere if you only need a lightweight slide generator or a static content pipeline: the interactive features and multi-agent runtime add complexity, have runtime costs tied to LLM/TTS providers, and benefit most from a deployment that handles real-time streaming and media.
Practical notes
The repo is designed for self-hosting (Node.js + pnpm) and Vercel/Docker deployment, but production use requires configuring at least one LLM/TTS provider or running the optional Lemonade local stack. The team validated the MAIC concept with real students at Tsinghua, and the project includes integrations (OpenClaw) to trigger classroom generation from chat apps for rapid demos and internal deployment workflows.
