Your crew & fleet

Your crew & fleet

Anvil isn't one assistant — it's a crew you delegate to, running across a fleet of machines you steer from one place. This is the map of how it fits together.

Anvil gives you more than a single assistant. You get a crew — specialists work is routed to, and persistent workers that stand by between sessions — running across a fleet of machines you manage together. The pieces have precise names, and it's worth getting them straight, because they do different jobs.

The four ideas

What it is
SquadA capability the orchestrator delegates to — "I need security work." You name the need; the host picks the specialist.
AgentThe specialist the host resolves from a squad to actually do a delegated task, in its own context.
BotA persistent worker on a machine's team — a named crew member with a role, a model, and a ceiling it can't exceed.
InstanceA machine running Anvil. Connect several to AnvilHub and they become a fleet.

Read it as a sentence: you keep a crew of bots; the orchestrator routes work by squad; the host runs the right agent; and all of it happens on one or more instances.

How delegation keeps context lean

The reason Anvil can have a large crew without drowning the model is squads: the model sees twelve capability names, not a roster of dozens of agents. It asks for a capability; the host resolves the specialist. That cost stays fixed no matter how many agents you add — and you can add a whole new squad by installing a marketplace package, with no Anvil release.

From one machine to a fleet

A single instance is complete on its own. Pair two or more to AnvilHub and you get a fleet: a view of every machine, the ability to hand a task to whichever one should run it, and a single place where all their work and cost rolls up. Fleet routing is one of the things that only works through AnvilHub — see Instances & the fleet for why.

Seeing it all

Three AnvilHub surfaces turn the crew from something you trust into something you watch:

  • Team — the bot board: who's on the crew and what each is doing now.
  • Runs — every run across the fleet, with its changeset, outcome and cost.
  • Instances — every machine, its status, and where your workspace is pointed.

Where to next