Summarise with AI action settings
Add Summarise with AI from Page actions → Add action after the source answers are available. It is an action (type="aiSummarise"), with normal trigger timing and conditions.
Shared AI input settings
These inputs also apply to Categorise with AI.
| Setting | Property | Behaviour |
|---|---|---|
| Text source | aiTextFieldId | Select one text field, or use a text template instead. |
| Text template | aiTextTemplate | Liquid producing the text to process, at most 64 KiB. |
| Uploaded images | aiImageFieldIds | File fields containing accepted image uploads. |
| Image URLs | aiImageUrlTemplates | Fixed URLs or Liquid templates, one per item, subject to integration destination restrictions. |
| Additional instructions | aiInstructions | Optional guidance, at most 8,000 characters. |
Text alone, images alone or both are supported; runtime input needs non-empty text or at least one image. Use either the text field or text template. The combined image limit is ten, across both source lists: PNG, JPEG or WebP, at most 10 MiB per image and 20 MiB in total. Oversized inputs fail rather than being silently clipped or resized.
Summary settings reference
| Setting | Property / choices |
|---|---|
| Summary format | aiSummaryFormat: paragraph (default) or bullets. |
| Summary length | aiSummaryLength: short (about 50 words), standard (about 150; default), or detailed (about 300). These are approximate targets, not exact limits. |
State the audience and facts to preserve in the instructions. For example: “Summarise the reported problem, location and impact. Preserve uncertainty. Do not infer a cause or promise a resolution.” Keep original answers available for review.
Read or save the result
{{ actions.summaryAction.response.summary }}Replace summaryAction with your action ID. Use Add result to form and select response.summary to save it into a calculated field. See Saved action results for independent applicant/reviewer visibility and failure handling.
Test AI and environment behaviour
User view returns labelled free mocks with isMock=true. Save settings, then use Test AI for a real call with separate example text/images. It runs no other actions, saves no answers and retains uploaded examples only for the request.
Real Test AI, QA, Staging and Production calls use the linked account's AI allowance and any agreed funded overages. Valid successful results are chargeable on metered accounts; failed AI calls are free. AI processing does not also count as an automation action. Contract/legacy unlimited accounts follow their account arrangements. Check current Usage and Overages and credits; Darcy design assistance has separate allowances.
Handle failures and verify the journey
AI has a total 15-second deadline including preparation/admission, with no automatic AI retry. Failures use the normal action error/response code; timeout is 504. Adding the action does not automatically require output or block submission.
Test blank input, real examples, long text, images, timeout/failure handling and source changes. Check names, numbers, dates, negation and uncertainty against the original. Use normal calculations and conditions for exact arithmetic and rules.