AICosmos
Educational multi-agent platform — incubated at Tsinghua, live at aicosmos.ai.
AICosmos turns AI from a tool into a teammate in the classroom — a multi-agent operating layer for education and research, incubated by Tsinghua University’s Department of Computer Science.
My role. Contributor — agent design and platform features. Status. Live at aicosmos.ai. Tagline. AI for Education × AI for Research — empowering teaching and discovery through agentic AI.
Why agents, not just chatbots?
Today’s AI in education is mostly a Q&A bot: it answers facts but doesn’t act with the learner. Two gaps remain:
- Disembodied action — AI can’t touch the real workflow, so knowledge stays inert.
- Decontextualization — knowledge is detached from social and physical context.
AICosmos closes both gaps by composing three classes of agents into a single platform:
| Agent | Role | What it does |
|---|---|---|
| Tutor Agent | Coach | Guides learners through reasoning steps rather than handing answers. |
| Role-Play Agent | Co-actor | Plays a specific role to make interaction situated, not abstract. |
| Q&A Bot | Reference | Answers factual questions on demand. |
Together they form an educational agent: an entity that can perceive a learning environment, plan autonomously, and act to support the teaching goal.
What’s underneath
- Multi-agent orchestration for course-specific workflows
- Tool use — agents can call into simulation, retrieval, and grading
- Profile-driven — adapts to a learner’s history and curriculum
- API-first — drops into existing teaching platforms
Links
- 🌐 Live: aicosmos.ai
- 🏛️ Incubated by Tsinghua University, Department of Computer Science
assets/img/aicosmos_cover.svg with the official logo / product screenshot when ready.