One office, one AI box: serving every workstation from a shared server
One capable machine can serve a whole office. AI Server binds the LAN behind an API key, the model loads once for everyone, and nothing leaves the building.
A single AI Server on the office's GPU box serves every workstation on the network over HTTP with API-key authentication; team apps point at it instead of running models locally [1]. One GPU serves twenty laptops because the model loads once and everyone shares it. The data stays in the building, there is no cloud AI in the path and no per-token bill, and IT pulls or updates models in one place rather than on every desk [1].
Keyed by default, fail-closed
Network serving is a paid capability; a loopback-only server stays free [3]. You issue an API key on the serving machine from the desktop app or the CLI; the raw key is shown once and the server stores only a hash. The daemon refuses to start on a non-loopback bind if no key exists, which is the fail-closed default rather than an error to work around [2]. Requests without a key on /v1/* return 401; health probes stay keyless so monitoring works.
Discovery that pins the server
Workstations find the box in Settings under AI Server: mDNS discovers it on the LAN, or you add http://<server-ip>:<port> and paste the key [2]. Each app pins the server's fingerprint on first connect, so a later fingerprint change warns before it is trusted rather than silently rerouting traffic.
Your other tools work too
Any OpenAI-compatible tool can use the same endpoint: base URL /v1, the API key as a bearer token [1]. Anonymous third-party traffic is rate-capped; keyed traffic is not. The office keeps one endpoint and one key policy instead of a separate integration per tool.
What IT operates
The server hosts its own dashboard: active clients, traffic attributed by app, and the installed model list, reachable from any desk with the key [1]. This is the step before a farm. When one box stops being enough, the same product moves behind a gateway without the apps changing.
References
- Software Tailor. "AI Server — product page." softwaretailor.com/ai-server.htm. Accessed 2026-07-22.
- Software Tailor. "AI Server documentation — Quick start." softwaretailor.com/docs/ai-server/quick-start.htm. Accessed 2026-07-22.
- Software Tailor. "AI Server documentation — Licensing." softwaretailor.com/docs/ai-server/licensing.htm. Accessed 2026-07-22.
Related articles
Serve the whole office from one box.
One GPU, every workstation, no cloud. Get it on Windows from the Microsoft Store, or read the feature tour first.