Each shape below uses the same product and the same licence key; you can move from one to the next without changing client code.
1. Developer workstation
Shape: AI Server on the developer's own PC, serving only that PC. Plan: Free.
Every AI Suite app and local script on the machine uses one set of models; nothing is exposed to the network. The right start for evaluation and for single-user automation.
Walkthrough: The zero-exposure default · Install on Windows
2. Office server
Shape: One GPU workstation serving the office LAN, as a Windows service, with an API key per person or app. Plan: Pro Personal for household and education use, Pro Commercial for a business.
Workstations stay light: staff use AI Client or AI Suite apps, which find the server on the network. The most common first deployment.
Walkthrough: One office, one AI box · Security model
3. Cloud or data-centre container
Shape: The softwaretailor/aiserver image on a VM or container host in your own cloud account, behind your load balancer with TLS. Plan: Pro Commercial.
Use it when the GPUs are in your cloud or data centre rather than an office. Health probes and drain-on-shutdown make it fit any orchestrator.
Walkthrough: Self-hosting AI in your own VPC · Run in Docker
4. Gateway farm
Shape: An AI Gateway in front of several workers, on Docker Compose or Kubernetes. Plan: Pro Commercial, one licence per worker.
One address for clients, failover, model-aware routing, canary releases and upgrades without downtime. Add workers as use grows; autoscale on Kubernetes.
Walkthroughs: AI Gateway farm · Model-aware routing · Sticky affinity · Capacity and backpressure · Canary rollouts · Zero-downtime upgrades · Deploy on Kubernetes
5. Branch office
Shape: A gateway appliance at each site; capable machines that join the farm when they are switched on, through local-network discovery with a fingerprint allowlist. Plan: Pro Commercial.
For clinics, shops and labs without IT staff on site: prepare boxes centrally, list their fingerprints on the gateway, and ship them.
Walkthrough: Branch-office AI
6. Regulated enterprise
Shape: A gateway farm with region policy, mandatory moderation, approved models only (curated mode), per-team quotas and budgets, and signed audit exports shipped to a SIEM. Plan: Pro Commercial, optionally enrolled in AI Admin Console.
For organisations that must show who used which model, prevent unapproved models, and keep evidence for auditors.
Walkthroughs: Regulated enterprise AI · Observability and chargeback · Governance
Also useful
- Legacy coexistence — old and new clients on one server during a migration, until the legacy API ends on 2026-12-31.
- Connecting an agent — proving a full tool-calling loop against your server.
Choosing
| If you need… | Start with |
|---|---|
| To try it, or AI for one person | 1 |
| AI for a team in one place | 2 |
| AI next to your cloud data | 3 |
| More than one GPU machine, or no downtime | 4 |
| Many small sites | 5 |
| Evidence for auditors and strict policy | 6 |
The deployment planner turns a workload into servers, seats and a price.