Agent
An Agent node runs one of your organization’s agents inside a workflow. The node holds only two things: which agent, and the Task to give it on each run. The agent’s record supplies the rest: its instructions, model, granted apps, skills and databases. A change saved to the agent applies to the next run of every workflow that uses it.
Agents in workflows covers this node in depth, with approvals and AI cost.
Settings
Section titled “Settings”| Field | What it does |
|---|---|
| Agent (required) | One of your organization’s agents. Its description shows below, with a link that opens its page. |
| Task | What the agent should do on this run. A template: type {{ to insert a variable. |
With no agents yet, the panel links to the Agents page to create one. If the picked agent is deleted later, the field reads “A deleted agent” until you pick another.
What happens in a run
Section titled “What happens in a run”- The step opens a new thread on the agent, titled after the workflow, the run and the node. The thread is listed on the agent’s Chat tab and links back to the run.
- It sends the rendered Task, after a fixed note that tells the agent no one will answer a question during the run. When the Task renders empty, it sends the text of the step before, or “Begin.” when there is none.
- The agent works with its own grants. When it calls an action that asks first, the step waits for a member to approve or deny it. Admins are emailed.
- The step finishes with the agent’s reply.
A call that waits 72 hours without a decision is denied with the reason “expired without a decision”, and the step fails. See Runs and approvals.
Output
Section titled “Output”| Field | What it holds |
|---|---|
response |
The agent’s reply |
thread |
The id of the run’s thread on the agent’s Chat tab |
tools |
Each tool the agent called, one per line, with the error if one failed |
approvals |
The app actions a member approved or denied during the step |
The next step’s {{ input }} is the reply. On the run page, Open thread in the step’s details
goes to the thread.
Publish checks
Section titled “Publish checks”Publish refuses an Agent node with no agent picked, with an agent that no longer exists, or with a Task that isn’t valid Liquid. The workflow templates leave every Agent node without an agent, so pick one in each before the first run.
AI cost
Section titled “AI cost”Each Agent step makes model calls on the agent’s provider. They run on your organization’s own AI Gateway key when one is set, or count against your plan’s AI allowance. See Models and AI keys.