DocsFeaturesSwarm Fleet
Multi-agent

Swarm Fleet

On larger tasks, Trace can split the work across several sub-agents running at the same time instead of doing everything in one long, sequential pass.

It's automatic, not a mode you switch on

There's no separate "swarm mode" toggle. When you ask for something and the agent decides the task genuinely benefits from parallel work — for example, several independent pieces that don't depend on each other — it spawns a small fleet of worker agents on its own, each handling one piece, coordinated by the same agent you're already talking to.

Watching it work

When a swarm spawns, you'll see a small live card appear in the conversation with a Swarm Monitor button. Click it to open a dedicated panel showing every worker agent, what each one is doing right now, and their progress — updated live as they work, not just once they finish.

Once every worker finishes, the coordinating agent picks the conversation back up automatically with their combined results — you don't need to do anything to "collect" the output.

You stay in one conversation
Even with several agents running underneath, you're only ever talking to one thread. The swarm is an implementation detail of how a single request got done faster — not a separate chat you have to manage.