Dashboard — /dashboard

Your home screen after signing in: a glance at the whole swarm, your pools, and what happened recently. Everything else is one click away in the sidebar, so the dashboard does not repeat it.

New pool in the page header is the page's one primary action. Admins with accounts waiting for approval also get a Review N pending accounts button beside it, which opens the users admin page.

At a glance

Five tiles across the top, computed from the pools you belong to (every pool, if you are an admin):

Tile	What it counts
Pools	Pools you are a member of; the line under it says how many you own, or how many more public pools you could join.
Nodes online	Online nodes across those pools, with how many are offline.
GPUs	GPUs on those nodes, with their total VRAM.
Models served	Models loaded and ready for chat, plus any still loading or failed.
Jobs running	Training runs in progress, plus how many are queued.

Node and GPU numbers come from each pool's health snapshot, which only members can read; a public pool you have not joined is listed but not counted. A tile you are not allowed to read, or that the leader could not answer, shows - and says Not available rather than a misleading zero.

Get started

A four‑step checklist that tracks your progress and offers the next action as a button:

1. Create or pick a pool — done once you have access to a pool.
2. Add a node — opens your pool's Nodes tab, where Add a node generates the one‑line join command.
3. Train a model — submit a job from the Training console.
4. Serve and chat — load a trained model on the Inference page.

A progress bar in the card's header shows how many steps are done. Once all four are, the card folds into a small Get started: 4 of 4 done chip under the page title; select it to reopen the list. Both the card and the chip have a dismiss button (×). Dismissing hides the checklist for good in this browser (it is remembered in local storage, so a private window or cleared site data brings it back).

Cluster

Every node of your pools, with every GPU and what it is doing right now. The head line sums it up - nodes online, GPUs in use, VRAM used of the total, and the tokens per second all served models are generating at this moment.

• Models being served — a chip per model running on your nodes: its speed now (or per reply over the last 15 minutes) and how many stages it is split into. The name links to the Inference page.
• Nodes — one card per node: host, pool, online / training / offline, its round trip to the leader, and each GPU with a badge saying what it does - Serving stage N, Training , or Idle - plus load and memory bars, temperature and power. A node with two GPUs serving one model stage shows one busy and one idle card: a served stage uses one GPU.

The card refreshes every five seconds while the tab is visible. The same data is at GET /api/v1/cluster/overview (?pool= narrows it to one pool) and in the MCP tool cluster_overview.

Pools

A tile for every pool you can see: name, visibility (public or private), node count against the plan's quota, member count, and your role. Select a tile to open the pool page; opening it does not change your active pool.

For pools you belong to, each tile also shows a health line: a status dot with words (All 3 nodes online, 2 of 3 online, No nodes yet), and the pool's health grade as a pill (Grade C · 79.4; hover it for what the score means). A pool with too little history to grade says Ungraded rather than showing a bad grade.

The dashed New pool tile at the end of the grid opens the same dialog as the header button: name, visibility, an optional description, and the pool's owner (you, or an organization you own or administer) and address. The address preview under the fields shows where the pool will live, for example /pools/you/thuringia, and flags a taken slug before you submit (see Organizations and addresses). Only the visibilities your plan allows are offered; with a single choice the dialog says which one applies instead of asking. Creating a pool takes you straight to its page, where Add a node lives.

When your plan's pool quota is used up, the line under the heading spells out the allowance (1 of 1 pool(s) on your plan · up to 2 node(s) each), the header button is disabled, and the dashed tile reads Pool limit reached.

Manage all (admins only) opens the pools admin page.

Recent activity

The latest eight things that happened across your pools, newest first: nodes joining, leaving, going offline or coming back, grade changes, role changes, and training runs being queued, started, finished or failing. Each entry links to its pool (and training runs to the Training console) and says how long ago it happened; hover the time for the exact timestamp. The full timeline of a single pool is on its Overview tab.

The pool events come from one call, GET /api/v1/pools/activity (see pool metrics), so only pools you are a member of contribute - a public pool you have not joined stays out of your feed. If the leader does not answer, the list stays empty rather than failing the page.