A Rack vision model describes images for a text model
Put SmolVLM2 on the Rack's Vision task and a local text model that cannot see gets a written description of every screenshot instead.
Eyes for a text-only worker
Models now lists SmolVLM2 2.2B. It downloads with the image projector it needs, both from pinned, checked files, and waits on the Rack until you put it on Vision, VLM.
When a tool returns an image, such as a screenshot, the vision model describes it and copies any text in it, and the working model reads that description.
Nothing lost if it fails
Leave Vision empty and images pass through as before. If the vision model cannot start, the image is kept as it was for that call.