Run your agent and read the trace

Run it

Your "Page summariser" has instructions, the Fast model, and the tools http.fetch and notify.email.

1. Open the agent and find Run now.
2. In Input (optional) type a public page address, for example https://example.com.
3. Choose Run agent.

You land on the run page. If the agent is paused, Run agent is disabled; the page shows a warning banner and you would choose Resume (then save) first.

Reading the header

At the top you see "Page summariser: run", the trigger and start time. Beneath it:

• A status badge. The statuses are Queued, Running, Waiting for your approval, Done, Failed and Cancelled.
• A Live label while the run is in progress.
• A summary: number of steps, tokens used and cost.
• Open agent and, while it is active, Cancel run.
• An Input box showing what you gave it.

The trace, step by step

The steps appear in order as they happen:

1. Thinking shows the agent's reasoning.
2. Calls http.fetch: the agent chose the tool. Choose Show details to see the exact address it sent.
3. http.fetch returned: the page content, clipped in the list. Show details shows the full result. If the fetch fails, the title says http.fetch failed in red and shows why.
4. Thinking again as it reads the result.
5. Needs your approval: the agent wants to use notify.email. The run pauses.

Press Approve on the card at the top, or open Approvals. Choosing Details first shows the exact email text. After approval the run continues: notify.email returned, then Answer, and a green Result box with the final output. The status becomes Done.

If you choose Deny instead, the agent is told the request was refused and finishes without sending.

What failure looks like

A run can end with Failed and a red message. Common reasons:

• No AI key: "Add your own AI provider key to use this feature."
• Budget reached (next lesson).
• The agent used up its step limit: the message says the run reached its limit of steps and was stopped.
• A tool failed and the agent could not finish.

Temporary problems are retried automatically a few times. Unexpected errors show a reference starting with ERR- that support can use.

Use the trace to debug

The trace tells you what really happened. When a result is wrong, scroll up and ask:

• Did it call the tool I expected, with the arguments I expected?
• Did the tool return what I thought?
• Did the instructions tell it what to do with that result?

Then fix the instructions, save and run again. Use Cancel run if you realise mid-run that something is wrong; it stops and shows Cancelled.

Try it yourself

1. Run the agent with a page address and approve the email.
2. Run it again, deny the email, and compare the two traces.
3. Run it with a page address that does not exist and read the failure.

Recap

• Run now starts a run and opens its page.
• The trace shows thoughts, tool calls, results, approvals and the answer.
• Show details reveals exactly what a tool received and returned.
• Approvals pause the run until you answer.