OpenMAIC 功能与快速上手：构建一个小型互动课程

Start with the smallest classroom

OpenMAIC 最小可验证单元是一个课堂会话：主题或材料进入工作台，代理规划活动，再生成可检查的课堂产物。先用一个短文档和一个学习目标，逐项验证材料导入、回答、产物、刷新后的持久化与导出。

The repository's documented baseline is Node.js 20+ and pnpm 10+. After cloning, install dependencies and copy `.env.example` to `.env.local`; configure one server-side model provider before adding media or search integrations.

Exercise the core features

A first pass should test four capabilities separately: material ingestion, an AI teacher or classmate response, one visual artifact such as slides or a whiteboard, and persistence after a page refresh. Then test an export such as editable PowerPoint or interactive HTML with a representative document.

OpenMAIC lists cloud and local options including OpenAI-compatible endpoints, Anthropic, Google, DeepSeek, Lemonade, and FunASR. Treat each provider as a separate capability matrix: chat, image, speech, search, and storage may need different credentials and failure handling.

Quickstart and failure diagnosis

Use `pnpm install`, create `.env.local`, and start the command documented by the release you pinned. If the UI loads but a lesson fails, check the browser request, server log, provider key, model name, and material size in that order. If export fails while chat succeeds, inspect the artifact renderer and file permissions rather than rotating API keys.

For a repeatable lesson, record the commit, provider, prompt, source-file hash, generated artifact paths, and human review decision. Never place provider keys in client-side code or commit `.env.local`.

A learning-oriented acceptance test

Call the run successful only when a reader can reproduce it, explain which agent or tool produced each artifact, and identify what still needs human review. Generated lessons are drafts until a subject-matter reviewer checks factual accuracy, accessibility, and age or audience appropriateness.