When Metir splits a job across agents

Some jobs are too big for one pass. Metir splits them across several agents working at once, and shows you all of them rather than hiding the work.

When it happens

You see a group when a larger request is divided into several clearly named pieces of work. The group opens while work is active and can be reopened after the answer finishes. For ordinary answer details outside a group, see reading a response.

What you see

A group in the conversation with one row per agent. Each row shows its state: Running, Completed, Failed or Cancelled. A running row also shows the current activity and elapsed time. A completed row keeps a short result line, so you can choose which one is worth opening.

Opening one

Select a row to open its detail dialog. The details do not expand inline inside the group. The dialog contains Prompt, Expected output, Tool calls and Report. Prompt shows the assigned request, Expected output shows the requested form, Tool calls lists the visible work, and Report contains the returned result. On mobile, the same row opens a full-height details sheet.

Checking the result

Start with Report, then inspect Tool calls when you want more context about how that row reached its result. A failed row keeps its error in the same detail view, while the other rows remain available. Source chips in the surrounding answer can still be opened using the approach in reading a research answer.

Why the work is shown

Because you can then see which agent produced which claim, instead of being handed a synthesised answer and asked to take it on trust. Keep the group with the conversation so you can return to its rows later. See history and search when you need to find that conversation again.