THU-MAIC/OpenMAIC：一键把任何主题变成多智能体互动课堂 THU-MAIC/OpenMAIC: Turn Any Topic into an Interactive Multi-Agent AI Classroom
OpenMAIC 是一个开源 AI 教学平台，先把主题或资料整理成课程大纲，再扩展为幻灯片、测验、互动网页和项目式活动；它强调多智能体课堂、深度互动、可恢复的 Pro 工作台和本地/服务端双部署路径。 OpenMAIC is an open-source AI teaching platform that turns a topic or learning materials into a course outline, then expands it into slides, quizzes, interactive web pages, and project-based activities. It emphasizes multi-agent classroom interaction, deep exploration, a recoverable Pro workspace, and both local and se
它解决什么问题
OpenMAIC 面向的是“把学习内容直接做成课”的场景。输入可以是文档、表格、图片、音频和视频，系统先分析材料生成课程大纲，再把每个条目扩展成幻灯片、测验、互动网页或项目式学习活动。对于快速搭建 AI 教学站点、课程导航页或内部培训工具，这意味着主题可以更快落成可交互课程，而不是停留在聊天问答。
课堂不是单聊天机器人
项目的核心体验是多角色参与：AI 老师和 AI 同学可以讲课、发起讨论、进行立场不同的圆桌辩论、回答问题，也可以通过语音讲解，在共享白板上写公式、画图和推导步骤。测验支持单选、多选和简答，并带有 AI 自动评分与反馈。项目式学习则允许学习者选定角色，和 AI 一起完成带里程碑和交付物的任务。
- 深度互动模式支持 3D 可视化、模拟实验、知识小游戏、思维导图和浏览器内在线编程场景。
- AI 老师可以直接操作互动界面，例如高亮重点区域、设置条件和给出提示。
- 这些互动页面面向桌面、平板和手机使用。
Pro 工作台与系统能力
v1.0.0 新增 Pro 工作台，用户可以直接用自然语言描述学习目标，Agent 会规划课程、创建和整理课程、读取或搜索课程 DSL、生成和修改页面，并编辑旁白与课件结构。它使用明确工具来插入、复制、删除、排序或补丁式修改页面，避免直接改动整块难以检查的数据。工作台还支持可恢复的持久会话，服务端重启后可以继续，运行中也能取消任务或追加指令。
- 内置 20 个技能，覆盖课程规划、深度研究、讲授、工作坊、职业教育、幻灯片制作、PPTX 导入、编辑和风格复用。
- 可创建自定义技能。
- 支持浏览器、HTTP、PostgreSQL 和 S3 存储后端，默认也能只保存在浏览器中。
What it solves
OpenMAIC targets the workflow of turning learning material into an actual class. Inputs can include documents, spreadsheets, images, audio, and video. The system first analyzes the material to generate a course outline, then expands each outline item into slides, quizzes, interactive web pages, or project-based learning activities. For AI teaching sites, course navigation tools, or internal training systems, that means a topic can become an interactive course faster instead of staying as a chat prompt.
This is not just a chatbot classroom
The main experience is multi-role participation. AI teachers and AI students can lecture, start discussions, run debates from different positions, answer questions, and explain via voice while drawing formulas, diagrams, and derivations on a shared whiteboard. Quizzes support single choice, multiple choice, and short answer, with AI grading and feedback. Project-based learning lets learners pick a role and complete milestone-driven tasks with AI.
- Deep interactive mode can generate 3D visualizations, simulation experiments, learning games, mind maps, and in-browser coding scenes.
- AI teachers can directly manipulate the interactive interface, such as highlighting key areas, setting conditions, and giving hints.
- These interactive pages are designed for desktop, tablet, and mobile use.
Pro workspace and system capabilities
Version 1.0.0 adds a Pro workspace. Users can describe learning goals in natural language, and the agent will plan lessons, create and organize courses, read or search the course DSL, generate and modify pages, and edit narration and slide structure. It uses explicit tools to insert, copy, delete, reorder, or patch pages instead of directly editing large opaque blobs of data. The workspace also supports recoverable persistent sessions, so work can continue after a server restart, and tasks can be canceled or extended while running.
- Built-in 20 skills covering lesson planning, deep research, teaching, workshops, vocational education, slide production, PPTX import, editing, and style reuse.
- Custom skills can be created.
- Supports browser, HTTP, PostgreSQL, and S3 storage backends, and can also run fully in the browser by default.
来源
- GitHub Trending · 08-31 17:27