AI SERVER & AI GATEWAY — DOCUMENTATION

AI Server documentation

AI Server turns one machine into a private AI platform: OpenAI-compatible APIs for chat, embeddings, images, speech and vision, served from hardware you own. Every AI Suite app discovers it automatically; AI Gateway mode scales one server into a load-balanced farm. These pages take you from first install to a production Kubernetes estate.

New to the product? Start with the AI Server product page for what it is and who it's for; come back here for the how.

Quick start

Desktop install, a single Docker container, and a Compose farm with the AI Gateway — the fastest routes to a working server.

Kubernetes & Helm

The Helm chart and kustomize manifests: workers, gateway, DNS discovery, autoscaling, GPU nodes, and the probe/drain contract that makes upgrades zero-downtime.

Licensing & activation

Tiers, activating desktop and headless/container nodes, how licensing behaves when our servers are unreachable, and offline operation.

Gateway & operations

Configuring the worker pool, load-balancing strategies, canary rollouts, the built-in dashboard, Prometheus metrics, and security hardening.

API & clients

The OpenAI-compatible endpoint reference, authentication, streaming, connecting generic tools and your own code, and the free-tier metering rules.

Need help?

Write to [email protected] or use the contact form (topic: Product / Support).

The 60-second mental model

One binary, three roles. The same server runs as a desktop-managed daemon, a headless container (AISUITE_ENGINE=real), or a gateway fronting a pool of workers (AISUITE_ENGINE=gateway).
Loopback is free; the network is paid. A server serving only its own machine needs no license or keys. Binding beyond loopback requires a paid tier and at least one API key — the daemon enforces both itself.
Clients don't change as you scale. A client talking to a single server talks to a gateway-fronted farm of twenty the same way — same URLs, same keys, same behaviour.
Everything is observable out of the box. Every node self-hosts a live dashboard at /dashboard and native Prometheus metrics — no agent or sidecar to install.