# S-Agent: Spatial Tool-Use Elicits Reasoning for Spatial Intelligence Project Page Paper arXiv Code Data --- Watch the S-Agent demo video on YouTube
## Overview **S-Agent** is a spatial tool-use agent for continuous multi-view image and video reasoning. It accumulates spatio-temporal evidence through spatial tools, scene memory, and agent memory, enabling vision-language models to reason over 3D space beyond a single visual impression. ## TODO - [x] Release paper and arXiv. - [ ] Publish the project page. - [ ] Open-source inference and evaluation code after the paper is public. - [ ] Release S-Agent trajectories / S-300K data after data cleanup. - [ ] Release model checkpoints and usage examples. - [ ] Add citation once the paper metadata is finalized.