Agent swarms

Run Fifty Agents at Once

Split a migration, a backlog, or a fleet-wide fix into one session per piece and start them all at once. Each agent gets its own sandbox, its own branch, and its own pull request, so nothing waits on anything else and one failure costs one session.

01

One list in, one session per item

A loop over your repositories or files is the whole fan-out: agent session start returns as soon as the session is created, so starting fifty takes seconds. Pass --watch on a second pass to block until each one finishes and exit non-zero on failure.

Placeholder: a shell loop fanning a repo list out into one session each

A shell loop turns a list into a swarm

02

Every agent works alone

Each session gets its own sandbox with the repositories cloned and the toolchain installed, so agents build and run tests without touching each other. Each opens its own pull request, reviewable and mergeable on its own.

Placeholder: parallel sandboxes, each with its own branch and pull request

One sandbox, one branch, one pull request per session

03

Watch the whole batch, and cap it

agent session list and the dashboard show every session in the batch with its status and cost. Each session runs under a hard budget, and trailing account, agent, and developer limits cap what the whole swarm can spend.

Placeholder: a batch of sessions listed with status and cost

Status and cost per session, under caps you set

You can use Ellipsis for...

Every use case runs on the same platform. See how it works