Appearance
Agentic Tasks
INFO
Agentic tasks will be released in version 2.42.0, planned for release on 21/10/2026.
An agentic task sends a request to an AI agent configured in AireGlu for your tenant, and stores the agent's response so later tasks can use it. Like other automated tasks, it starts as soon as it's created and finishes as Completed or Failed. Agentic tasks appear on the Tasks page alongside other task types.
Agentic tasks must be enabled for each tenant. Contact support@airelogic.com to have them enabled for yours. Until then, the Agentic task type isn't available in the designer, and agentic tasks aren't created when triggered.
Fields
| Field | Description |
|---|---|
| Agent | The AireGlu agent to use. Leave it on Auto to let AireFlow choose one from your tenant's agents. |
| Capability | The capability to use on the selected agent. Leave it on Auto to let AireFlow choose. You can only pin a capability if you've also pinned an agent. |
| Intent | Shown when Agent or Capability is on Auto. Describes what the task should do, such as Book an appointment for the patient, and is used to choose the agent and capability. |
| Input | Key-value pairs sent to the agent. Values can use XPath and Liquid. |
| Output | The fields the agent must return. Each value is optional, and can use XPath and Liquid. |
| Due date and expiry | As for other tasks. See Task Due Date and Task Expiry. |
Input and Output keys must start with a letter and can only contain letters, numbers, underscores and dashes, such as appointmentDate or appointment-date. Keys are case-insensitive, so Date and date count as duplicates.
Pin both Agent and Capability when you know which one the task needs, and use Auto when the right one depends on the request.
Output fields
If you set Output fields, AireFlow keeps asking the agent until every field has a value, up to 5 times. Each request tells the agent which fields are still missing.
- If an Output field's value isn't empty once mapped, it's used for that field and the agent isn't asked for it.
- An Output field with the same key as an Input value takes the Input value.
- If a request doesn't fill in any more fields, AireFlow stops early rather than repeating it.
The task fails if any Output field is still missing at the end. The workflow logs show which fields were missing.
Without Output fields, the agent is called once and its response is stored as it is.
Using the agent's response
When the task completes, its emitted context automatically includes an agentic element, so tasks triggered by it can use the response:
| Expression | Value |
|---|---|
{{ context.agentic.data.fieldName }} | An Output field, or a field from the agent's JSON response |
{{ context.agentic.summary }} | The agent's text response, if it returned one |
{{ context.agentic.dataJson }} | All of the response fields as a JSON string |
{{ context.agentic.routing.agentName }}, {{ context.agentic.routing.capabilityId }} | The agent and capability that were used |
For example, with an Output field of appointmentDate, a later HTTP task could send:
liquid
{ "appointmentDate": "{{ context.agentic.data.appointmentDate }}" }The response is also kept in the task's state, under agentic.output, and is viewable via AireFlow's Tasks page.
Retry on failure
Retry on Failure works the same as for other automated tasks. If it's set to Yes and the task fails, the whole task is run again a number of times before it's left as Failed. This is separate from the repeated requests used to fill in Output fields.