Let an Agent read images

A user can send a screenshot, photo, or chart to an Agent in chat. If the normal turn model cannot accept images, Ankole uses the model in the vision_fallback profile to read the image.

You can leave this profile empty when primary already supports image input. You can also leave it empty for an Agent that never receives images.

Configure an image model

1. Open Console → Agents and select the Agent.
2. Find vision_fallback under Model profiles.
3. Select a Provider and model that explicitly support image input.
4. Save the profile, then start a new conversation.

The model list does not determine image support for you. Check the Provider documentation before you select a model.

Verify the configuration

Send a clear image in a chat that is connected to the Agent. Ask a question that requires the image, for example:

What error does this screenshot show? Tell me the most likely cause.

If the Agent ignores the image or says that it cannot read it, check these items in order:

1. The chat channel uploaded and forwarded the image.
2. vision_fallback is saved on the correct Agent.
3. The selected model supports image input.
4. The Provider credentials and quota are available.

To let an Agent create images, see Image generation. Image generation and image understanding use different model profiles.