Show HN: Visual Agents with Code Mode

Category: ai-ml

Tags: vision-language-models, computer-vision, structured-data-extraction

Score: 6.3/10 (Innovation: 6, Technical: 7, Documentation: 6, Utility: 6)

VLM Run Cookbook provides Jupyter notebooks demonstrating structured visual understanding using Vision Language Models, with Orion 2 introducing code execution for complex computer vision tasks. Its innovation lies in combining deterministic tool-calling with dynamic code generation, making visual agents more reliable and inspectable for production use.

Target audience: data engineers, AI/ML engineers, backend devs

Repository: https://www.vlm.run/blog/orion-2 · Jupyter Notebook · 308 stars

View on Hacker News