Cages

A cage is an isolated Fargate container where your AI agent runs. Each cage has its own filesystem, network namespace, and persistent storage — no inbound ports, configurable egress.

What you get

• Full Linux environment — Ubuntu-based container with root access
• Persistent storage — EFS volume mounted at /workspace, survives restarts and hibernation
• No inbound ports — Nothing can connect to your cage directly (webhooks go through the gateway proxy)
• Configurable egress — Outbound internet access for API calls, package installs, etc.
• Environment variables — KMS envelope-encrypted, injected at boot

Cage sizes

Naming

Cage names must be unique per account. They’re used as identifiers in the CLI and API:

lobster deploy my-agent       # create cage named "my-agent"
lobster ssh my-agent          # connect to it
lobster status my-agent       # check its state

Lifecycle states

Unique per account

Each cage name is unique within your account. You can have multiple cages running simultaneously, up to your plan’s limit.