Run a custom agent
Give an agent a task from Runs on its page, follow its work in a chat session, and reread earlier runs.
Runs, on the right of the agent's page, is where you give the agent a task yourself and reread its earlier runs. Every run is a chat session: you see the agent's steps and answer, follow up, and download what it delivers. You need permission to view agents, by default only the Admin role, and chat access to start a run.
4 steps · About 3 minutes
Open Runs
Open the agent and click Runs on the right. Every run is a chat session with the agent. At the top, pick the profile the agent works for in this run; the menu lists the linked profiles. Switch on Send output email to email the result to the agent's output recipients. Earlier runs sit as tabs above, and + starts a new one.

Give the agent a task
Type a task in the box at the bottom, such as Write the report for week 45, and press Enter. Or click Run now to run the instructions as written. The agent shows its steps and its answer as it works. Tools that normally ask for approval ask you here too. Ask a follow-up in the same run to change or continue the result.

Choose the AI model
Every run uses the AI model in the agent's Settings: Auto, which picks the model and effort for each message, or one model with an effort of Low, Medium or High. Runs from the chat, a schedule or content use it too. For a run you start on Runs, the model menu in the task box picks another model for that run only.

Reread an earlier run
Click a tab at the top of Runs to open an earlier run, here Run. The list holds the runs you started here, runs from a tag in the chat and scheduled runs, newest first. You read the task and the answer, and you can follow up in the same run. Run again runs the instructions once more in that session.

FAQ
Run now is grey, or I can only read runs.+
Runs says You can view these runs. Running the agent or replying needs chat access in this workspace. Ask an admin for a role that may start chats.
Do I need to switch anything on before I run the agent here?+
No. Runs you start here don't depend on the switches under Gets work from. Those only decide whether the agent takes work from the chat, a schedule or content. See Let an agent work on its own.
What does Send output email do?+
It emails the final reply and the files the agent delivers to the output recipients. Leave it off when you are only trying the agent. Scheduled runs email the recipients on their own. See Manage agent output recipients.
The agent stopped before it did everything.+
A run has a limit on the number of steps it takes. On its last step the agent always answers with what it has so far, so a run never ends without a reply. Ask a follow-up in the same run to let it continue, or split the task into smaller ones.
I don't see Runs.+
Runs is on the page of every agent you may view. By default only the Admin role may view agents. To start runs, you also need chat access in the workspace.