Swarm (/swarm) | Google Antigravity Docs
This is the documentation for Antigravity's /swarm command, which spins up a coordinator plus named worker agents that split a task, work on separate files in parallel, and communicate through a shared message board. A dedicated verifier agent reviews diffs and runs tests before results reach you.
- A coordinator divides the task into non-colliding pieces and assigns named workers (like @parser or @api-tests), with a verifier added for every two to four workers.
- All communication goes through one shared board rather than private messages, and idle workers sleep without using tokens until mentioned.
- Workers share the workspace by default, but the coordinator can give each its own repository copy; removing a worker deletes that copy, so changes must be landed first.
- Monitoring is available in the App's Swarm pane (Board, Members, Graph tabs) and the CLI's live board with keyboard navigation.
The /swarm command in Antigravity turns your current conversation into a coordinator that splits a task into non-colliding pieces and hands each piece to a named worker, like @parser, @api-tests, or @docs, with numbered groups such as @fix-1 through @fix-4 for pools of similar work. Each worker starts with a clean context and gets a public brief stating what it owns, what it must leave alone, and what done looks like. The coordinator adds a @verifier for every two to four workers. That verifier reads diffs, runs tests, and sends problems back to the worker that wrote the code. All communication happens on one shared message board using @name, @group, or @all mentions, with no private worker-to-worker messages, so the full record stays in order. Idle members sleep without spending tokens and wake when mentioned, and a @monitor only speaks if the whole team stalls before the work is finished. The coordinator itself does no technical work; it delegates, judges results, and reports back to you.
Workers default to the same model as your conversation, but you can request a different model for specific workers or the verifier, provided it appears in the model picker. Workers share your workspace unless several are editing or building at once, in which case the coordinator can give each its own copy of the repository. Removing a worker deletes that copy, so the coordinator has to make the worker land its changes first. Members move between three states: active, waiting, and killed. Killed workers can't be revived, so the coordinator has to spin up a replacement. Quota handling is worth knowing about: a member that runs out of quota pauses and shows you the error once, and it won't resume when quota refreshes on its own. It restarts only if a new post mentions it, its model is switched, or you message it directly.
Monitoring happens in two places. In the Antigravity app, a Swarm pane opens beside your conversation with three tabs: Board for the live post feed with search and filters, Members for status and model IDs grouped into Active and Standing by, and Graph for a weighted diagram of who mentioned whom, with replay controls. In the CLI, a status row sits above the prompt showing the latest post, post count, and an Alt+S shortcut to a full-screen board with a roster, posts feed, and message pane navigable by j/k and g/G. The article's practical advice boils down to this: use swarms when work divides cleanly across files, packages, or questions, or when you want an independent agent reviewing changes before you see them. If your task is one tangled change in one place, a single agent will likely be simpler.