Reproduce “Vision-as-a-Tool: Visual Harness for Multimodal Agentic Coding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haolin Yang, Wenshan Wu, Jin Jiang, Yu Cheng, Yan Xia, Ting Song, Yangyu Huang, Zhiliang Peng and 8 more
The authors introduce VaaT, implemented as VHarness, which lets coding agents call an external multimodal model for visual feedback during code-and-render refinement. Across Image-to-HTML, Image-to-SVG, and Image-to-3D, four multimodal agents improve on reported metrics; text-only GLM-5.2 completes all three tasks with VHarness. Results cover static targets and do not establish faster or lower-token completion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.