Media

Image support

SEND steps support inline image attachments, allowing your AI Agent to deliver visual context alongside text in Playbook responses. Use images for troubleshooting guidance, product visuals, instructional diagrams, or any scenario where a picture reduces confusion and chat abandonment.

How to add images

1. In a SEND step, type @ on your keyboard to open the inline menu.
2. At the top of the menu, under Media, select Add image.
3. In the image picker, reuse an existing image from the Search tab (see Reuse images), or upload a new file from the Upload tab.
4. When uploading, add alt text (required for accessibility). A reused image keeps the alt text it was given when it was first uploaded.
5. The image appears as an inline attachment in the step body.

Images are delivered deterministically — the Agent sends the exact image you select at authoring time, not AI-generated content. Multiple images per SEND step are supported and delivered in sequence.

Reuse images

The image picker keeps a library of the images already uploaded to an Agent’s Playbooks, so a given image only has to be uploaded once and can be reused across Playbooks and steps.

In the Search tab of the picker:

• Browse the library of previously uploaded images, or filter it by typing part of an image’s alt text.
• Select one or more images, then insert them together — each is added as its own inline attachment, in the order selected.
• Images already used in the Playbook being edited are marked In use. They can still be selected and inserted again.

The Search tab shows and filters the 100 most recently uploaded images, so once an Agent has more than 100 images the older ones can’t be found by searching. The library also only lists images that at least one Playbook still references — once the last reference is removed, the image stops appearing in the Search tab and can no longer be picked for reuse.

A newly uploaded image only becomes available to reuse once the Playbook it was added to is saved. Until then it won’t appear in the Search tab.

Reused images point at a single stored file rather than a per-Playbook copy. Deleting a Playbook, or removing the image from it, never deletes that shared file — other Playbooks that use the image keep working, and images already sent in past conversations still display. So removing an image from a Playbook is always safe: at worst it removes the image from the picker (if that was the last Playbook using it), which affects only whether you can pick it again — not any Playbook that already has it. Editing an image’s alt text, however, updates the shared library entry that every author sees.

Dynamic images from a Variable

An image in a SEND step can be backed by a Variable instead of a fixed upload, so one Playbook can deliver a different image per conversation.

To insert an image from a Variable:

1. In a SEND step, type @ to open the inline menu.
2. Under Media, select Insert image from variable — listed alongside Add image.
3. Pick a Variable with the Media data type. Only Media Variables are listed.

Alt text is not entered for these images. It is set from the Variable’s name at the moment you insert it — hero_image becomes “Hero image” — and cannot be edited afterward. Renaming the Variable later does not update the alt text already stored on the image; remove the image and insert it again to pick up the new name.

At runtime, the Variable’s value replaces the image URL. Media Variables hold a publicly reachable http or https image URL. If the Variable is empty, or holds a value that is not an image URL, the attachment is skipped and the rest of the step is still delivered.

Media Variables are populated by a SET step — which can assign a URL or clear the value — or by an API tool that writes its output to the Variable. They cannot be captured from the end user: the Agent cannot receive an image upload into a Variable.

Because the URL is substituted at delivery time, the image never passes through the LLM. The Agent sends exactly the image the Variable points to.

Images from an API tool in AI-generated replies

When an API tool returns an image URL into a Media Variable, an AI-generated reply that refers to that Variable delivers the image itself rather than a link the end user has to open. This covers cases where the right image isn’t known at authoring time — a product photo for the order the end user is asking about, for example.

Only the exact value stored in a Media Variable is turned into an image. Other links in the same reply — Knowledge article links, handoff links, URLs inside code formatting — remain text.

Delivery matches the channel support table above.

No authoring step is required for this. Wherever the stored value appears in the finished reply, it is delivered as an image. Alt text comes from the Media Variable’s name.

Alt text can also now be generated for an image hosted on your own systems rather than uploaded to the Agent. The image must be served over https and be a PNG, JPEG, or GIF.

Supported formats

JPEG, PNG, and GIF (including animated) are supported. Each image must be 5 MB or smaller; oversized files are rejected at upload time.

Accessibility

Alt text is required when uploading an image. Alt text is included in conversation transcripts and read by screen readers, ensuring end users who rely on assistive technology receive equivalent information.

For an image backed by a Media Variable, alt text comes from the Variable’s name instead and is not editable — so give those Variables descriptive names before inserting them.

Channel support

Images are supported across the following channels:

Modes

Images are supported in both SEND modes — Fixed message and Contextual AI-generated.

Handoff visibility

Images sent by the AI Agent during a Playbook are preserved after a human agent handoff. Human agents can view them in Salesforce Live Chat (Ada Glass), Salesforce Case Creation, Zendesk Live Chat, Zendesk Ticketing, and Zendesk Messaging.

Known limitations

• The AI Agent cannot process or understand images sent by end users.
• Knowledge Base images are not supported.
• Videos and other file types are not supported.
• The LLM does not review images for safety or confidentiality before sending.
• Carousels are not supported. Multiple images are delivered as separate attachments in sequence, not as a single multi-image message.
• Media Variables cannot be captured from the end user, used in conditions, or created through the Platform API.