Run a custom agent in chat
Tag a custom agent with @ in a chat message, add the task, and follow its run and report in one card.
In any workspace chat you can give a custom agent a task: type @, choose the agent, write the task and send. A card in the conversation follows the run and ends with the agent's report. You need permission to view agents and to chat; by default only the Admin role can view agents.
3 steps · About 3 minutes
Open a chat
In the sidebar, click Chat and open a conversation, or click New conversation. You give an agent a task from the message box at the bottom, next to your own questions.

Tag the agent and add the task
Type @ and part of the agent's name. The list marks agents with Agent and shows their tagline; only agents with Tag in the chat switched on are listed. Choose one: it becomes a chip in your message. Write the task after it, such as Write the report for week 44, and send. You can tag one agent per message.

Follow the run card
After you send, a card shows the agent and what it is doing, such as Currently: and the tool it uses. When it is done, the card says Run complete and shows the report. Chips show what the agent got: fields Growf took from your message, such as Week, and its settings with a lock. Ask a follow-up in the chat to discuss or reuse the report.

FAQ
The agent isn't in the @ list, or it doesn't start.+
The list only offers agents with Tag in the chat switched on, under Gets work from in the agent's Settings. When a setting marked Required on the agent's Input and output is empty, the chat tells you to fill it in first. See Let an agent work on its own.
The card says it is waiting for a field.+
The agent needs a value for a required per-task field that Growf could not find in your message. Fill it in on the card and click Start run, or click Cancel.
Can I tag two agents in one message?+
No. You can tag one agent per message. Send a second message to give another agent a task.
Where else can I see the run?+
On the agent's page, under Runs. The run shows up there as a tab, and you can follow up with the agent directly. See Run a custom agent.
Can I ask about the data behind the agent's answer?+
Yes. Ask a follow-up in the same chat. The chat can look up the saved tool calls, results, errors, task and report of the run without running the agent again.
Which model and tools does the agent use here?+
The AI model from the agent's own Settings, not the model of your chat. Its tools run without asking for approval, with the profile you tag or the connections of the workspace. The run is not emailed to the agent's output recipients.