Built-in tools›Agent orchestration

Group

Agent orchestration

Turned on by

Its own switch on the Tools page of Conversation settings.

Approval

Reviewed: writes and open-ended actions ask for approval unless the security level or an earlier “Always allow” covers them.

Profile key

tool.fork.description

Asks you to start a new, separate conversation from a task the model writes; nothing from that conversation comes back here.

The new conversation starts with the model's text as its only message and none of this conversation's history. To get an answer back instead, the model uses agent_spawn.

When it asks you

Every call puts a card, The model wants to fork this conversation, in the top-right corner of the window, at every security level. The model does not wait for your answer and is never told it, so you can answer later.

When you approve, the new conversation starts running under the source's top-level conversation in the sidebar, with the source's settings, workspaces and worktree. An unanswered card expires after 30 minutes, and quitting Mework drops it.

Good to know

• This is not Fork conversation in the timeline's right-click menu, which copies the history above the line into a new conversation.

Parameters