Data fields

Data fields are the result an agent gives back from every task, in the same fields each time. When a task finishes, they are filled in from what it did, then:

• shown as a table under the task's summary,
• shown as columns on the agent's Tasks list, so you can compare results across tasks,
• returned by the API, from the task's structured response.

A value the task did not find stays empty; it is never guessed. A field can combine values from the task, such as the total of several invoices.

Set them up

1. Open the agent and click Data fields.
2. Click Add a field for each value. Give it a name such as invoice_number, a type (Text, Number, Date, Yes / no, List or Group) and, if the name is not obvious, a description.
3. Under How to fill them, add anything about format, for example "dates as DD/MM/YYYY".
4. Click Save version.

The agent works from its role and steps, not from the data fields, so make sure they cover the values you want. An agent asked only to "read the invoice" may never look for the due date.

Data fields need AI, so automations do not have them.

From the API

A run started from the API can ask for different fields for that run with output_config, which replaces the agent's data fields. See the API reference.

Last updated on