AI Actions

The AI group is new since 0.14.2. It covers two different jobs: generating parts of a scenario from a description, and extracting structured data from unstructured content.

note

These actions send content to the Aiviro AI service. Do not point them at data your organization is not allowed to process externally, and use Result is secret / Sensitive Data on the surrounding actions when handling personal data.

Prompt Actions

A container action that builds its own child actions from a plain-language description. Describe what should happen, and the AI generates the actions inside the block; you then review, edit and re-order them like any hand-built section.

Because the generated actions stay visible in the Commands tree, the AI is only a starting point — nothing runs that you have not seen.

📸 Screenshot needed — /img/editor/actions/ai/prompt_actions.png

Capture: a Prompt Actions block with a Title, a filled-in Prompt and several generated child actions expanded underneath it in the Commands tree. Why it is new: this action did not exist in 0.14.2.

Parameter	Description
Title	Label shown on the block in the tree.
Prompt	Description of what the generated actions should do.

tip

The block remembers the prompt the current children were generated from. Re-running generation after editing the prompt replaces the children, so keep hand-tuned actions outside the block — or copy them out before regenerating.

AI Read

Extracts structured fields from a document, a screen, the clipboard or a variable. Where Get reads one value from one place, AI Read takes a whole page of unstructured content and returns the fields you asked for.

📸 Screenshot needed — /img/editor/actions/ai/ai_read.png

Capture: the AI Read form with Source Type set to File, Instructions filled in, and two or three rows in the Extract template table showing field names and their types. Why it is new: this action did not exist in 0.14.2.

Parameter	Description
Source Type	Where the content comes from: Screen (what the robot currently sees), File (a path you supply), Clipboard, or Text (the contents of a variable).
Instructions	What to extract, in plain language — e.g. "Extract person's age, name and address from the document."
Extract template	The fields to return, each with a name and a data type. This defines the shape of the result.
Save to variable	Variable that receives the extracted data. Default processed_text.
Send image context	Also sends the image to the model, which improves results on documents where layout carries meaning — tables, forms, invoices. Default off.
Custom JSON schema	An explicit JSON schema for the result, for cases where Extract template is not expressive enough. Optional. (Advanced options)

tip

Name the fields in Extract template the way you want to use them later — the result is a structure keyed by those names, so invoice_total reads better in the rest of the scenario than field_2.

Process Invoice

Reads an invoice PDF and returns its parsed content — header data and line items — without you defining extraction rules.

📸 Screenshot needed — /img/editor/actions/ai/process_invoice.png

Capture: the Process Invoice form, and ideally the Variables panel showing the parsed structure it produced. Note: the old image is actions/files/pdf_reader.png — the action was renamed from Aiviro PDF Reader and moved from Files to AI, so the screenshot needs both a new name and a new capture.

Parameter	Description
PDF File Path	Invoice PDF to read.
Save to variable	Variable that receives the parsed result.

---

Which one should I use?

Task	Action
Read one known value from a known place on screen	Get (Tools)
Read several fields from a document or a busy screen	AI Read
Read a standard invoice	Process Invoice
Pull fixed values out of a PDF with positional rules	Extract PDF custom (Files)
Draft a section of the scenario for you	Prompt Actions