Documentation Index
Fetch the complete documentation index at: https://docs.affzero.com/llms.txt
Use this file to discover all available pages before exploring further.
What it does
The AI Extract step reads an attachment (typically a PDF) or a text variable and extracts a set of fields you define — such as invoice number, vendor name, amount, and due date. Each extracted field becomes a variable available in all subsequent steps.
This step is most commonly used in Invoice Capture automations triggered by the Email Received trigger, but it works with any source that provides a PDF attachment or text data.
Configuration
Choose what the AI reads:
| Option | When to use |
|---|
| Attachment from trigger/prior step | Read a PDF (or other file) attached to the inbound email |
| Text from a variable | Read a text variable, e.g. {email_body_text} from the trigger |
Attachment (when source = Attachment)
Select which attachment to extract from: first, second, or third. Attachments are numbered starting from 1 and come from the inbound email trigger.
Variable name (when source = Text variable)
Enter the variable name to read, e.g. email_body_text.
Output variable prefix (optional)
A prefix applied to every extracted field name. For example, if you set the prefix to invoice and you have a field named amount, the output variable will be {invoice_amount}.
Using a prefix is strongly recommended when running multiple AI Extract steps in the same automation, to avoid variable name collisions.
Define the schema — the list of fields the AI should look for. For each field:
| Setting | Description |
|---|
| Field name | Lowercase, underscore-separated name (e.g. invoice_number, amount). Becomes part of the output variable name. |
| Type | One of: Text, Number, Currency, Date |
| Description | A hint that helps the AI understand what to look for (e.g. “Total amount due including tax”). More detail = better accuracy. |
| Required | If checked and the field can’t be found, the step will fail. |
Output variables
Each field you define produces one output variable. With a prefix of invoice and fields number, vendor, amount, date:
| Variable | Example value |
|---|
{invoice_number} | INV-2024-0042 |
{invoice_vendor} | Acme Corp |
{invoice_amount} | 1250.00 |
{invoice_date} | 2024-03-01 |
- Write descriptive field descriptions. The AI uses them as hints. “Total amount due including VAT” gives better results than just “amount”.
- Use Currency type for monetary values. The AI normalises values like
$1,250.00 or € 1.250,00 to a plain number.
- Use Date type for dates. Output format is
YYYY-MM-DD.
- Avoid required fields if the document might not include them. Non-required fields that aren’t found are simply left empty.
- Works best with digital (text-based) PDFs. Scanned or image-only PDFs may produce empty results — multimodal (image) extraction is planned for a future release.
Example: Invoice Capture
A typical Invoice Capture automation uses AI Extract like this:
- Email Received trigger (inbound email with PDF attachment)
- AI Extract — extracts
number, vendor, amount, currency, date, due_date with prefix invoice
- Save to Drive — saves the PDF with filename
{invoice_vendor}_{invoice_number}.pdf
- Write to Google Sheets — logs a row with
{invoice_number}, {invoice_vendor}, {invoice_amount}, {invoice_date}
See Invoice Capture preset for a guided setup wizard that pre-configures all of these steps.
