Developer Builds Open-Source Tool to Add Low-Cost Vision to Text-Only AI Agents

A developer has released an open-source project called Free Vision Skill, designed to give text-only AI coding agents the ability to process visual information without switching to a more expensive multimodal model. The tool introduces a concept called Visual Evidence Packet (VEP), which compresses image content into only the task-relevant facts — such as error messages, file paths, or chart values — rather than generating lengthy full descriptions. This approach keeps token usage low and prevents the vision model from overstepping into reasoning or decision-making, which is left to the main text model. The tool supports local caching of vision results to avoid redundant API calls when the same image is processed multiple times. The project is aimed at developers who want to keep their existing text-based agent workflows intact while adding on-demand visual input as a modular, replaceable layer.