Agones Autopilot

Bring the Agones Fleets, GameServers and FleetAutoscalers you already have. No cluster to run, and servers nobody is playing on sleep.

Agones Autopilot is in alpha. It is real and usable; expect rough edges, and tell us about them through Support.

Agones Autopilot runs the Agones manifests you already apply to a Kubernetes cluster — without the cluster. There are no nodes to size, no versions to choose and no Agones to install. Your game server keeps calling the Agones SDK, and your matchmaker keeps calling the Agones allocator.

Agones Autopilot → Apply manifests → paste fleet.yaml → Check → Apply

What it accepts

Kind	What it becomes
`Fleet` (`agones.dev/v1`)	A fleet of identical game servers, rolled onto a new template the way Agones rolls one
`GameServer` (`agones.dev/v1`)	A single game server of its own
`FleetAutoscaler` (`autoscaling.agones.dev/v1`)	A `Buffer` policy keeping a fleet's Ready servers topped up
`GameServerAllocation`	Not applied — allocate through the allocator or `plany agones autopilot allocate`

Manifests are taken as written. Anything Autopilot cannot run the way a cluster would — a second container, a volume, a Static port — is refused by name when you apply, never quietly dropped. Fields that only mean something on a cluster, such as a fleet's scheduling or a container's imagePullPolicy, are accepted and listed back to you as warnings. The full list is in Using Autopilot.

Images can come from Plany's Container Registry (.plany.cr) or from any public registry.

How it differs from a cluster

Servers nobody is playing on sleep. A Ready server nobody has connected to for ten minutes goes to sleep and stops costing compute. It stays allocatable, and an allocation or a player connecting wakes it — the first connection waits a few seconds while it resumes exactly where it was.
Players connect to an address that does not change across a sleep and a wake. Allocations hand out that address and port.
A project and region replace a cluster and namespace. metadata.namespace is kept as a label, and names are unique per kind within a project and region.
One container and one port per game server. See what is supported.
There is no Kubernetes API. You apply manifests with the CLI, Terraform, the console or the API — not with kubectl — and the game server cannot call the Kubernetes API from inside.

Pricing

You pay for a game server's compute only while it is awake.

	Price
vCPU (`agones-autopilot-vcpu-hour`)	$0.0208 per vCPU-hour, while awake
Memory (`agones-autopilot-gib-hour`)	$0.0091 per GiB-hour, while awake
Local storage (`agones-autopilot-local-storage-gib-hour`)	$0.000182 per GiB-hour
Its address (`playpute-proxy-hour`)	$0.005 per hour, per game server

That is Playpute's compute price plus 30%. A sleeping server costs its storage and its address, not its CPU or memory — so a buffer of two Ready servers waiting for players is cheap until somebody plays on them.

Each game server runs at the smallest machine type that fits its container's resources; see Sizing.

Everything draws on your account's prepaid credit.

Limits

Quota	Hobby	Pro
Agones Autopilot fleets per project per region	3	50

Game servers also count against the project's ordinary instance, vCPU and memory quotas, like any other game server.