Image Vision (Reading Images)

The Image Vision ability lets an agent understand images you attach — even when the agent's own model is text-only. It's the read side of images; making images is the separate Image Generation ability, so you can grant one, the other, or both.

Two ways an agent handles an image:

• Quick look (delegate). The agent calls process_image with a specific question (e.g. "what colours are in the top-right logo?"). Behind the scenes that question and the image go to a vision-capable model (the one ticked Image-in in Models & Providers), and the answer comes back — all without the main agent leaving its own model. Want more detail? It just asks again with a sharper question.
• Take over (switch model). For an image-heavy conversation (say a web-design review), the agent can call switch_model to run itself on a model that can both see images and use tools. That choice sticks for the conversation until it switches back. See Choosing & Switching Models.

Switching needs your permission by default (you can set it to automatic in the agent's Tools panel). If no image-reading model is configured, the agent will tell you to tick Image-in on a vision-capable model in App Config.

Related: Abilities, Attachments & Voice Input, Image Generation, Delegation & Hand-off.