AI text steps

These four actions make one request to the AI and hand the answer to later steps.

Action	You give it	It returns
Sort into a category	Text, and a list of categories	category (one of yours, spelled as you wrote it) and confidence (0 to 1)
Pull out details	Text, and a list of details to find	One value per detail, named as you wrote it. Empty when the text does not say.
Summarise	Text	text: a few plain sentences
Write or answer	An instruction	text: what the AI wrote, with no preamble

Read a result in a later step as {{steps.<step id>.<value>}}, for example {{steps.sort.category}}.

Typical use: sort an incoming message with Sort into a category, then use a Switch on steps.sort.category to send each kind down its own path.

If Sort into a category answers with something that is not one of your categories, the step fails with "AI answered ..., which is not one of the categories". Make the categories distinct and few.

Read a file

Each one-shot step has an optional File setting. Give it a link to a file and the AI reads the file along with your text.

File type	How it is read
Pictures	As an image.
PDFs	Page by page, as images.
Spreadsheets (.xlsx, .csv)	The first sheet, as text.

Rules:

• The link must be to a file in your Cashfin media. The usual source is {{fileurl}} from the File uploaded trigger. Any other address is refused with "the file to read must be a link to a file in your Cashfin media".
• Other file types are refused with "AI steps can read pictures, PDFs and spreadsheets (.xlsx, .csv); this file is not one of those".
• A file that is too large is refused with a message giving the limit.