Running an Agent

A published Agent runs inside an ordinary Task, alongside your conversation and files.

Starting a run

In any Task, type / in the message box and choose the Agent, then attach the inputs it asks for. The run pins the Agent's current published version, so later changes to the Agent never affect it.

Following the workflow

While the run works, its workflow shows each step's state: not started, running, waiting for you, done or stopped. Checks show whether they passed or failed, and only the branch the workflow took is highlighted. The activity record beside it lists what the run did.

When the run needs something from you, such as a missing input or a review, it pauses and asks in place. Answering resumes the same run.

LNG Buyer Brief

Version 2

Waiting for approval

Arranging workflow…

Select a step to see how it works

Run the October edition with this source pack

• Japan customs imports, September.xlsx
• Korea terminal schedules.pdf
• Taiwan tender notices.pdf
• JKM and TTF prices, September.csv

Waiting for you

Waiting for you · 6 updates›

Waiting for you

Axe is holding this task until the step you started elsewhere is finished. It resumes on its own once that happens.

Cancel task

Reading the result

A finished run leads with what it delivered:

1. Files it produced, ready to open.
2. Figures, the short answers such as a rating or a ratio, each with its calculation and sources.
3. Passages, the longer text, lists and tables, each with its sources.

An output the run could not establish is marked as missing, with the reason, such as not found or access blocked. See Evidence and sources.

Correcting and repeating

If a result is wrong, choose Correct in Teach. It opens the Agent's draft with the run and its outputs as context, so your correction becomes part of the method. Publish again to use it in future runs.

Every run of an Agent is listed on the Runs tab of the Agent's page in Agents, with its status, version and result.