You can let Heptabase AI read your images in the following ways:

New chat / Add to chat reuses the existing image from the whiteboard. Pasting an image directly into the chat stores it as a separate chat image attachment, so it creates an additional image record without adding another image to the whiteboard. To avoid storing another copy, use New chat / Add to chat.

​

Mentioning a card does not guarantee that the Agent has read every image inside it. When the Agent reads the card’s content, embedded images can be included for image processing, subject to the selected model’s image support and the limits below.

​

To draw attention to a specific image, add that image directly and say what you want the Agent to inspect.

​

The following are the image limits Heptabase applies to each model request, grouped by provider. They are not the providers’ universal limits for every product. The selected model must support image input. Sizes in the table are in MiB.

​

File sizes here use MiB (1 MiB = 1,024 × 1,024 bytes). The limits apply to an individual request, not the total number of images stored in your Space.

​

Anthropic requests with more than 20 images: each image must have a longest side of 2,000 pixels or less. Images that exceed this size, or whose dimensions cannot be determined, are omitted from that request.

​

A card reference does not automatically expand all linked or embedded content into the request. Images also need to meet the model’s supported formats, size, count, and any pixel-dimension limits. Some images may be omitted, or the request may fail, when those conditions are not met.

​

Try a model that supports images, resize the image, send fewer images, or split the task into smaller requests. For a particular image, attach it directly through one of the supported routes above.