Run an agent
Start an agent on demand from its page by clicking Run. If the agent asks for inputs, the Run agent window opens so you can add documents, answer its questions, and add instructions for this run. Agents without inputs start straight away. Anyone the agent is shared with, including people with view-only access, can run it.
Start the run
Runs start from the agent's own page.
Open the Agents page and select the agent.
Click Run.
If the Run agent window opens, fill it in as described below and click Run.
August opens the run page, where you can follow each step on the run graph. The run also appears in the agent's Run History.
To run the agent inside a chat instead, open the menu next to Run and choose Run in Assistant.
Add documents
Document inputs offer two ways to add files, and you can combine them:
Upload files adds files from your computer. You can also drag files onto the input.
Add from folders opens a picker for files and folders already in August. Click Done to add your selection.
Remove any item from the list before running. Run stays unavailable while uploads are still in progress.
Review inputs and context
The window shows what the run will use:
Saved defaults applied to every run: the documents, playbooks, tabular reviews, skills, and agents saved with the agent. They are included automatically.
Inputs: the questions the agent asks for each run, such as a counterparty name or a choice from a list. Required inputs must be filled in before you can run.
Extra instructions (optional): context or instructions for this run only.
Agent email tasks create or edit Outlook drafts by default and do not send. If agent email sending is enabled for your firm and the agent has a send grant for the specific recipient addresses, the run can send to those recipients, and an approval card may appear for you to confirm the send. Sends are limited to the granted recipients; a send to any other address is refused. Without both, the run produces a draft in Outlook rather than sending.
If the run pauses
A run that reaches a Question step shows as Waiting until someone replies. See Resume an agent run waiting on human input for how to reply and how long a run can wait.
What to read next
Agents Overview for high-level agent concepts
Agent Triggers for scheduled and upload-triggered runs
Build and edit agents on the canvas for steps, branches, outputs, and builder controls