OpenMAIC — Multi-Agent Interactive Classroom

Skill by ara.so — Daily 2026 Skills collection.

OpenMAIC (Open Multi-Agent Interactive Classroom) is a Next.js 16 / React 19 / TypeScript platform that converts any topic or document into a full interactive lesson. A multi-agent pipeline (LangGraph 1.1) generates slides, quizzes, HTML simulations, and project-based learning activities delivered by AI teachers and AI classmates with voice (TTS) and whiteboard support.

Project Stack

Layer	Technology
Framework	Next.js 16 (App Router)
UI	React 19, Tailwind CSS 4
Agent orchestration	LangGraph 1.1
Language	TypeScript 5
Package manager	pnpm >= 10
Runtime	Node.js >= 20

Installation

bash

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

Environment Configuration

bash

cp .env.example .env.local

Edit

.env.local

— at minimum one LLM provider key is required:

env

# LLM Providers (configure at least one)
OPENAI_API_KEY=$OPENAI_API_KEY
ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY
GOOGLE_API_KEY=$GOOGLE_API_KEY

# Recommended default model (Gemini 3 Flash = best speed/quality balance)
DEFAULT_MODEL=google:gemini-3-flash-preview

# Optional: MinerU for advanced PDF/table/formula parsing
PDF_MINERU_BASE_URL=https://mineru.net
PDF_MINERU_API_KEY=$MINERU_API_KEY

# Optional: access code for hosted mode
ACCESS_CODE=$OPENMAIC_ACCESS_CODE

Provider Config via YAML (alternative to env vars)

Create

server-providers.yml

in the project root:

yaml

providers:
  openai:
    apiKey: $OPENAI_API_KEY
  anthropic:
    apiKey: $ANTHROPIC_API_KEY
  google:
    apiKey: $GOOGLE_API_KEY
  deepseek:
    apiKey: $DEEPSEEK_API_KEY
  # Any OpenAI-compatible endpoint
  custom:
    baseURL: https://your-proxy.example.com/v1
    apiKey: $CUSTOM_API_KEY

Running the App

bash

# Development
pnpm dev
# → http://localhost:3000

# Production build
pnpm build && pnpm start

# Type checking
pnpm tsc --noEmit

# Linting
pnpm lint

Docker Deployment

bash

cp .env.example .env.local
# Edit .env.local with your API keys

docker compose up --build
# → http://localhost:3000

Vercel Deployment

bash

# Fork the repo, then import at https://vercel.com/new
# Set env vars in Vercel dashboard:
#   OPENAI_API_KEY or ANTHROPIC_API_KEY or GOOGLE_API_KEY
#   DEFAULT_MODEL (optional, e.g. google:gemini-3-flash-preview)

One-click deploy button is available in the README; it pre-fills env var descriptions automatically.

Lesson Generation Pipeline

OpenMAIC uses a two-stage pipeline:

Stage	Description
Outline	AI analyzes topic/document and produces a structured lesson outline
Scenes	Each outline item is expanded into a typed scene: `slides` , `quiz` , `interactive` , or `pbl`

Scene Types

Type	Description
`slides`	AI teacher lectures with TTS narration, spotlight, laser pointer
`quiz`	Single/multiple choice or short-answer with AI grading
`interactive`	HTML-based simulation (physics, flowcharts, etc.)
`pbl`	Project-Based Learning — choose a role, collaborate with agents

API Usage — Generating a Classroom

REST: Start Generation Job

typescript

// POST /api/generate
const response = await fetch('/api/generate', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    topic: 'Quantum Entanglement',
    // Optional: attach document content
    document: markdownString,
    // Optional: model override
    model: 'google:gemini-3-flash-preview',
  }),
});

const { jobId } = await response.json();

REST: Poll Job Status

typescript

// GET /api/generate/status?jobId=<jobId>
const poll = async (jobId: string) => {
  while (true) {
    const res = await fetch(`/api/generate/status?jobId=${jobId}`);
    const data = await res.json();

    if (data.status === 'completed') {
      console.log('Classroom URL:', data.classroomUrl);
      break;
    }
    if (data.status === 'failed') {
      throw new Error(data.error);
    }
    // status === 'pending' | 'running'
    await new Promise(r => setTimeout(r, 3000));
  }
};

REST: Export Slides

typescript

// GET /api/export/pptx?classroomId=<id>
const exportPptx = async (classroomId: string) => {
  const res = await fetch(`/api/export/pptx?classroomId=${classroomId}`);
  const blob = await res.blob();
  const url = URL.createObjectURL(blob);
  // trigger download
  const a = document.createElement('a');
  a.href = url;
  a.download = 'lesson.pptx';
  a.click();
};

// GET /api/export/html?classroomId=<id>
const exportHtml = async (classroomId: string) => {
  const res = await fetch(`/api/export/html?classroomId=${classroomId}`);
  const html = await res.text();
  return html;
};

OpenClaw Integration

OpenMAIC ships a skill for OpenClaw, enabling classroom generation from Feishu, Slack, Discord, Telegram, etc.

Install the Skill

bash

# Via ClawHub (recommended)
clawhub install openmaic

# Manual install
mkdir -p ~/.openclaw/skills
cp -R /path/to/OpenMAIC/skills/openmaic ~/.openclaw/skills/openmaic

Configure OpenClaw

Edit

~/.openclaw/openclaw.json

jsonc

{
  "skills": {
    "entries": {
      "openmaic": {
        "config": {
          // Hosted mode — get access code from https://open.maic.chat/
          "accessCode": "$OPENMAIC_ACCESS_CODE",

          // Self-hosted mode — local repo + server URL
          "repoDir": "/path/to/OpenMAIC",
          "url": "http://localhost:3000"
        }
      }
    }
  }
}

OpenClaw Skill Lifecycle

Phase	What Happens
Clone	Detect existing checkout or clone fresh
Startup	Choose `pnpm dev` , `pnpm build && pnpm start` , or Docker
Provider Keys	Guide user to edit `.env.local`
Generation	Submit async job, poll, return classroom link

Custom Scene Development Pattern

Scenes are typed React components. To add a new scene type:

typescript

// types/scene.ts
export type SceneType = 'slides' | 'quiz' | 'interactive' | 'pbl' | 'custom';

export interface CustomScene {
  type: 'custom';
  title: string;
  content: string;
  // your fields
  metadata: Record<string, unknown>;
}

typescript

// components/scenes/CustomScene.tsx
'use client';

import { type CustomScene } from '@/types/scene';

interface Props {
  scene: CustomScene;
  onComplete: () => void;
}

export function CustomSceneComponent({ scene, onComplete }: Props) {
  return (
    <div className="flex flex-col gap-4 p-6">
      <h2 className="text-2xl font-bold">{scene.title}</h2>
      <div dangerouslySetInnerHTML={{ __html: scene.content }} />
      <button
        className="mt-4 rounded-lg bg-blue-600 px-6 py-2 text-white"
        onClick={onComplete}
      >
        Continue
      </button>
    </div>
  );
}

Multi-Agent Interaction Modes

Mode	Trigger	Description
Classroom Discussion	Automatic	Agents proactively start discussions; user can jump in or get called on
Roundtable Debate	Scene config	Multiple agent personas debate a topic with whiteboard illustrations
Q&A	User asks question	AI teacher responds with slides, diagrams, or whiteboard drawings
Whiteboard	During any scene	Agents draw equations, flowcharts, or concept diagrams in real time

MinerU Advanced Document Parsing

For complex PDFs with tables, formulas, or scanned images:

env

# Use MinerU hosted API
PDF_MINERU_BASE_URL=https://mineru.net
PDF_MINERU_API_KEY=$MINERU_API_KEY

# Or self-hosted MinerU instance (Docker)
PDF_MINERU_BASE_URL=http://localhost:8888

Without MinerU, OpenMAIC falls back to standard PDF text extraction.

Supported LLM Providers & Model Strings

typescript

// Model string format: "provider:model-name"
const models = {
  // Google (recommended)
  geminiFlash: 'google:gemini-3-flash-preview',   // best speed/quality
  geminiPro: 'google:gemini-3.1-pro',             // highest quality

  // OpenAI
  gpt4o: 'openai:gpt-4o',
  gpt4oMini: 'openai:gpt-4o-mini',

  // Anthropic
  claude4Sonnet: 'anthropic:claude-sonnet-4-5',
  claude4Haiku: 'anthropic:claude-haiku-4-5',

  // DeepSeek
  deepseekChat: 'deepseek:deepseek-chat',

  // OpenAI-compatible (custom base URL)
  custom: 'custom:your-model-name',
};

Export Formats

Format	Endpoint	Notes
PowerPoint `.pptx`	`GET /api/export/pptx?classroomId=`	Fully editable slides
Interactive `.html`	`GET /api/export/html?classroomId=`	Self-contained HTML page

Common Patterns

Generate a Classroom from a Document String

typescript

const generateFromDocument = async (markdownContent: string, topic: string) => {
  const res = await fetch('/api/generate', {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      topic,
      document: markdownContent,
      model: process.env.DEFAULT_MODEL ?? 'google:gemini-3-flash-preview',
    }),
  });

  const { jobId } = await res.json();

  // Poll until done
  let classroomUrl: string | null = null;
  while (!classroomUrl) {
    await new Promise(r => setTimeout(r, 4000));
    const status = await fetch(`/api/generate/status?jobId=${jobId}`).then(r => r.json());
    if (status.status === 'completed') classroomUrl = status.classroomUrl;
    if (status.status === 'failed') throw new Error(status.error);
  }

  return classroomUrl;
};

Check Provider Health

typescript

// GET /api/providers
const checkProviders = async () => {
  const res = await fetch('/api/providers');
  const { providers } = await res.json();
  // providers: Array<{ name: string; available: boolean; models: string[] }>
  return providers.filter((p: { available: boolean }) => p.available);
};

Troubleshooting

Problem	Solution
`No LLM provider configured`	Set at least one of `OPENAI_API_KEY` , `ANTHROPIC_API_KEY` , or `GOOGLE_API_KEY` in `.env.local`
Generation hangs at outline stage	Check API key quota; try switching to `google:gemini-3-flash-preview` for higher rate limits
TTS not working	TTS requires a browser with Web Speech API support; check browser console for errors
PDF parsing produces garbled text	Enable MinerU by setting `PDF_MINERU_BASE_URL` in `.env.local`
Vercel timeout during generation	Increase function timeout in `vercel.json` ; generation is async so the API should return a `jobId` immediately
Docker build fails	Ensure `DOCKER_BUILDKIT=1` and that `.env.local` exists before running `docker compose up --build`
OpenClaw skill not found	Run `clawhub install openmaic` or manually copy `skills/openmaic` to `~/.openclaw/skills/`
`pnpm install` fails on Node < 20	Upgrade Node.js to >= 20 ( `nvm use 20` )
Port 3000 already in use	Set `PORT=3001` in `.env.local` or run `PORT=3001 pnpm dev`

Key File Structure

OpenMAIC/
├── app/                    # Next.js App Router pages & API routes
│   ├── api/
│   │   ├── generate/       # POST lesson generation, GET status
│   │   ├── export/         # pptx / html export endpoints
│   │   └── providers/      # LLM provider health check
│   └── classroom/          # Classroom viewer pages
├── components/
│   ├── scenes/             # Slide, Quiz, Interactive, PBL components
│   ├── whiteboard/         # Real-time whiteboard rendering
│   └── agents/             # Agent avatar & TTS components
├── lib/
│   ├── agents/             # LangGraph agent graph definitions
│   ├── providers/          # LLM provider abstractions
│   └── generation/         # Outline + scene generation pipeline
├── skills/
│   └── openmaic/           # OpenClaw skill definition
├── server-providers.yml    # Optional YAML provider config
├── .env.example            # Environment variable template
└── docker-compose.yml      # Docker deployment config

openmaic-classroom

NPX Install

Tags

SKILL.md Content

OpenMAIC — Multi-Agent Interactive Classroom

Project Stack

Installation

Environment Configuration

Provider Config via YAML (alternative to env vars)

Running the App

Docker Deployment

Vercel Deployment

Lesson Generation Pipeline

Scene Types

API Usage — Generating a Classroom

REST: Start Generation Job

REST: Poll Job Status

REST: Export Slides

OpenClaw Integration

Install the Skill

Configure OpenClaw

OpenClaw Skill Lifecycle

Custom Scene Development Pattern

Multi-Agent Interaction Modes

MinerU Advanced Document Parsing

Supported LLM Providers & Model Strings

Export Formats

Common Patterns

Generate a Classroom from a Document String

Check Provider Health

Troubleshooting

Key File Structure