One containerized AI Server runs on a cloud VM or any Docker host, configured entirely by environment variables, with orchestrator-grade health probes and JSON logs <a href="#ref-1">[1]</a>. Deploys are immutable and repeatable: the image is the artifact and a mounted volume carries the state. Every knob is an environment variable, so nothing is baked into the image, and <code>/livez</code> and <code>/readyz</code> mean a load balancer or orchestrator manages it correctly rather than guessing <a href="#ref-2">[2]</a>.
The image is the artifact, the volume is the state
The container image build is reproducible from the repository, and per-deployment state lives on a volume rather than inside the image <a href="#ref-2">[2]</a>. That separation is what makes a rebuild safe: you replace the image without touching the model store, keys, or licence state. Structured JSON logs (<code>AISUITE_LOG_JSON=1</code>) drop straight into a log collector.
Probe-ready for orchestrators
<code>/livez</code> reports the process is alive; <code>/readyz</code> reports it is ready to take traffic and flips to <code>503</code> while draining <a href="#ref-2">[2]</a>. Wire the liveness probe to <code>/livez</code>, the readiness probe to <code>/readyz</code>, and set a termination grace period at or above the drain and shutdown budgets, and rolling updates become zero-downtime without extra tooling.
Licensed and keyed before it serves
Because the container binds a non-loopback address, that is network serving, and the daemon fail-closes without both a Pro entitlement and at least one API key on the volume <a href="#ref-3">[3]</a>. A licence artifact is a per-install file written by a licensed install, not something to hand-author; a Free install is meant to be loopback-only by design.
The building block of a farm
A gateway farm is N of these containers behind one gateway, so the same env-configured image scales from a single VPC endpoint to a load-balanced fleet <a href="#ref-1">[1]</a>. The container that serves your mobile and browser heads today is the worker you replicate tomorrow.