Dashboard Widgets — The SETI Stack

The SETI Stack is SETI AI's visual analytics surface: a freeform canvas where you compose your own operations dashboard from 40+ drag-and-drop widgets. Every widget is live — there is zero mock data in the data path. A widget either shows a real reading from your hardware and services, or it tells you plainly that it is waiting for data. Nothing is invented to make a screenshot look busy.

This page explains the widget catalog, how live data reaches each widget through the AnalyticsBus, and how an operator builds and saves a dashboard.

Related: Power Analytics · Inverter Drivers · Energy Nodes

What you get

• A blank canvas, your layout. You start with an empty board and drag in only the widgets you care about. Pan and zoom across an infinite canvas, drag each widget by its header, and your arrangement is saved automatically.
• Live by default. Widgets poll or stream from real endpoints (/api/victron, /api/system, /api/analytics/*, /api/notifications/*). Refresh cadences are tuned per widget — fast-moving solar telemetry refreshes every 5–10 seconds, cost analytics every few minutes.
• One consistent design language. Gold-accented, dark, dense, Font Awesome icons throughout — the same look across the Command Center, Node Editor, and Setup surfaces.

How a user composes a dashboard

1. Open the Analytics Palette. The left sidebar lists every widget, grouped into collapsible sections, with a search box ("Search widgets...") to filter by name.
2. Add a widget. Either drag a palette item onto the canvas (drop where you want it) or click it to spawn it at a default position. Widgets are transform-pan and zoom aware, so a dropped widget lands exactly under the cursor regardless of how far you have panned.
3. Arrange. Drag any widget by its header to reposition. Mouse deltas are scaled by the current zoom level so dragging always tracks the cursor.
4. Bind data-bound widgets (see below) by typing a cluster ID and choosing a metric in the small picker row at the top of the widget body.
5. It saves itself. Every change — add, move, remove, or re-bind — calls saveCanvasState(), which persists the layout both to the browser (localStorage) and to the server (POST /api/canvas/save). Reload the page and your board, including each widget's binding choices, comes back exactly as you left it.
6. Export / import / clear. The Quick Actions widget and sidebar footer let you export the layout as JSON, reload a saved layout, refresh every widget on the board at once, or clear the canvas.

Because layouts are saved server-side as named canvases, the same dashboard can be recalled across sessions and devices.

AnalyticsBus — the live data backbone (zero mock data)

Data-bound widgets never fetch on their own schedule and never fabricate a number. They subscribe to a shared, single-source-of-truth event bus called the AnalyticsBus:

widget → AnalyticsBus.subscribe(cluster, metric, callback)

The bus delivers updates from three coordinated sources so a widget is both instant and always current:

1. Live push (SSE). A Server-Sent Events stream at /api/analytics/stream pushes a telemetry event every time a workflow writes a new value (/api/pipeline/output). Bound widgets update the moment new data lands.
2. Poll fallback (30s). A background 30-second poll against /api/analytics/data keeps widgets fresh even if the SSE connection drops (the bus auto-reconnects after a 5-second backoff).
3. Immediate first value. On the very first subscribe, the bus does a one-shot fetch (?hours=24) so the widget shows a real reading right away instead of an empty box.

The bus caches the last payload per cluster|metric key and fans it out to every subscribed widget, so ten widgets watching the same metric cost one data feed, not ten.

Binding controls — point a widget at a data source

Every data-bound widget renders a binding row at the top of its body: a free-text cluster ID field plus a metric dropdown (Voltage, Current, Temp °C, SOC %, SOH %, Power). When you change either, the widget transparently unsubscribes and re-subscribes to the new (cluster, metric) tuple, and the choice is written into the widget's saved options (node.opts) so it persists across reloads.

This is the bridge between the Node Editor and the dashboard: any workflow that POSTs to /api/pipeline/output for a given cluster + metric will light up every widget bound to that pair — live over SSE, or within 30 seconds via poll.

The "no fake data" guarantee in practice. If a widget has nothing real to show, it says so — "awaiting data", "No data yet — collecting…", "No temperature sensor data available", or a driver_pending notice — rather than rendering a random curve. The backend enforces the same rule: energy nodes return {driver_pending: true} when a real reading is not available.

Widget categories

Widgets are organized by category, each with its own accent color. The four DATA families (KPI, Gauge, Chart, Status) are the composable, bind-anything widgets that ride the AnalyticsBus. The rest are purpose-built panels for solar, battery, sites, system health, and operations.

The catalog

DATA · KPI — single-value readouts (bindable)

DATA · GAUGE — dials and bars (bindable)

DATA · CHART — time-series (bindable)

DATA · STATUS — state and history (bindable)

SOLAR / ENERGY — live power surfaces

The three solar widgets — Energy Flow, Victron Solar, and Power Overview — are also packaged as reusable modules in the widget registry (web/widgets/) so they can be embedded across both the dashboard and the analytics canvas from a single source of truth.

BATTERY — pack health

PIPELINES / OPERATIONS / SITES / MONITORING

SYSTEM — infrastructure health (real Docker/host data)

Why it matters

• Build the view your role needs. A site operator can put Energy Flow, Inverter Hub, and Active Alerts front and center; an analyst can fill a board with bound charts and KPIs tied to specific battery clusters. The layout is yours and it persists.
• Trust every pixel. Because the AnalyticsBus is the single source of truth and the backend refuses to fake readings, a number on the SETI Stack is a real number from real hardware or a real service — never a placeholder.
• One canvas, live end to end. Workflows in the Node Editor publish to clusters; widgets on the dashboard subscribe to them. Data flows from your inverters and batteries through the pipeline and onto the screen with no manual refresh.