Qwen Office

A team at the National Astronomical Observatories, Chinese Academy of Sciences (NAOC) — its Xinglong Observing Station and Galactic 3D Structure group — used Qwen Office's Agent Harness and MCP interface to build a research-grade telescope simulation environment in under three days for about 1,000 yuan (~$140). It can rehearse equipment failures and multi-target competition, scheduling cases that are hard to recreate on real instruments.

The team's earlier StarWhisper Telescope (星语望远镜), built on Qwen, puts observation planning, telescope control and data processing in one agent framework. Stanford HAI's AI Index Report 2026 cites it as a representative 2025 AI agent for physics, astronomy, chemistry and materials science. The simulation chain now links to the real transient follow-up pipeline for early candidate assessment of supernovae and gamma-ray bursts.

Qwen Office helped a team at National Astronomical Observatories, Chinese Academy of Sciences that included Li Yuyang build a research-grade digital simulation system for a large-aperture telescope in three days for less than $143 (under 1,000 yuan), compared with an earlier outsourced approach that took about three months and cost tens of thousands of yuan (several thousand dollars).

Telescope components, sensor states and observation environments are integrated as standard MCP interfaces, letting an agent run the closed loop of state awareness, task planning, plan generation and process verification on the Sitian Pathfinder (司天探路者) and Sitian prototype (司天原型机). The system has flagged eight very early supernova candidates, two of which triggered follow-up observations, and the simulation data will be used to train the telescope control model VLA.