The editor

The editor is one screen: a canvas with a bar at the top, a bar at the bottom, and two floating buttons.

The canvas

Gesture	On empty canvas	On a node card
Tap	Select a wire, or clear the selection	Select the node
Drag	Pan	Move the node
Hold	Draw a marquee to select several	Enter multi-select
Two fingers	Pinch to zoom, drag to pan	Same — a pinch that starts on a card still zooms

Zoom controls and a Fit to screen button sit in the corner, and the zoom level is shown as a percentage.

A node’s ports are the handles down its two sides. Drag from a handle to another handle to wire them. Dragging from a port onto empty canvas opens the palette restricted to nodes that can connect there — a suggestion, not a rule, with a Show all escape hatch.

The top bar, in two modes

With nothing selected the bar is about the workflow: back, its name, the Enabled switch, and an overflow menu (rename, duplicate, delete…).

With something selected it is entirely about the selection: how many nodes, configure, duplicate, delete, and a ✕ to clear.

They never share the bar. That is why the name has room to be read and why “1 node selected” is spelled out in full.

Adding nodes

Add node opens the palette. It is grouped by node kind — Triggers, Actions, Values, Transforms — with each kind’s categories collapsed underneath, and a search box at the top that matches names and descriptions. Searching for hue finds the light nodes even though their type ids say light; searching for whatsapp finds the messenger nodes.

A node card

Every card carries, left to right:

• execution inputs — the socket that says run me next;
• data inputs — one per wired value, coloured by type;
• the node’s icon, name and a line of its own status;
• data outputs, then execution outputs.

Tap a card to open its config sheet. Each field is a form row; a field marked @Wired has a small socket toggle beside it that turns the row into a data input port on the card. A field that holds an identifier — a place, a variable, an app, a sound, another macro — opens a chooser rather than accepting typed text, for the reason in Identifiers are chosen.

A card also shows what it is missing: a permission the phone has not granted, hardware the phone does not have, or a reference to something deleted. Where a permission can be granted, a Grant button is on the card itself.

The bottom bar

Three surfaces, each replacing the canvas rather than covering it.

Problems is what the graph is. Errors and warnings, each naming the node or the wire it is about, with a badge on the bar itself so you do not have to open it to know there is something there. It is not a log — see Validity.

Console is what a run did. Every node that ran, every value that crossed a wire, every console.log from a script, every failure. Filterable by All / Info / Problems, and a row taps through to the whole entry with a copy button and a “go to the node” link. See The console.

Variables is what the state is right now — this workflow’s declarations with their live values, and a row through to the shared globals.

Running by hand

There is no whole-graph Run button, deliberately. A Manual Trigger node carries its own Run button on its card — which becomes Stop while that run is going — so a macro full of other triggers can still be started by hand, and an armed macro can be run without disarming it.

The same path serves the home-screen widgets and launcher shortcuts — see Widgets and shortcuts.

The AI assistant

A small button above Add node opens the assistant. It is the only surface that sits over the canvas rather than instead of it, because its edits land live: nodes appear as they are made and the canvas follows whatever was just touched, until your first pan or zoom hands control back for the rest of the turn.

The panel folds by dragging its handle — the conversation collapses, the input never moves. ↺ starts a new conversation and leaves the graph alone; ✕ closes the assistant.

It needs an AI connection and a model profile. See AI.