User Guide
The User Guide walks through every feature of Server AI Hub in the order you'll encounter them.
:::note Status Each page below is a stub. Content fills in over the next few weeks. Pull requests welcome at github.com/MARKYMARK55/ServerAIHubManual. :::
Dashboard
The main view. Live GPU, memory, disk, swap, network, and container health. Click any gauge to deep-link into the relevant Grafana dashboard.
Services & Models
Browse the catalog of LLMs, embedders, judges, and rerankers. One-click load with the recommended container settings; or roll your own via the Launcher.
Finder
The graphical file manager. Browse /home, /4tb-drive, mounted network shares. Upload, download, rename, share, search. Right-click for context actions (open in terminal — for the rare 1% — copy as cURL, etc).
Terminal
The escape hatch. xterm.js bound to a real bash inside the dashboard container, with file-tab autocomplete and persistent history. You should not need this often; it exists for the rare cases.
Containers
Every container running on the box. Start, stop, restart, view logs, edit compose. Width toggle (full / half / auto) to fit your screen.
LiteLLM
The model router. See which models are live, their routing weights, and their cost-per-token. Add a new endpoint (Hub-local or cloud) with a form, no YAML.
Launcher
For the day you want to roll your own — Sparkrun, vLLM, Ollama, your own image. The Launcher generates the compose file and surfaces it back to the Containers page.
Backup
Borg snapshots + iCloud sync. Pick what to back up, how often, where to. Restore from the same UI.
Telemetry
Health summary, container metrics, model latency, spend, alert history. Everything is also in Grafana — Telemetry is the curated index.
Admin
Operational status. Antivirus state, firewall events, audit log, host info, users, storage, packages, services. The "what's happening right now" view.
Settings
How the box is configured. Network, endpoints, notifications, API tokens, LLM costs, schedules, sudo, GitHub. The "change how things work" view.