Pipelines

A pipeline is a directed graph of nodes that moves and transforms data. Where the namespace answers "what is this value?", a pipeline answers "what should happen to it?"

One trigger, one read, one calculation, two destinations.

When to use a pipeline

Situation	Use
Show a live value on a screen	A binding — no pipeline needed
React to a button press on a screen	An event action
Move data on a schedule or an event, reliably	A pipeline
Transform, enrich, aggregate before storing	A pipeline
Expose an HTTP endpoint other systems call	A pipeline (REST ingress)
Bridge one protocol to another	A pipeline, or a second binding for simple cases
Server-side logic invoked from a screen	A Gateway script

The distinguishing property of a pipeline is reliability: writes it performs are covered by store-and-forward — buffered on outage, replayed in order, quarantined on permanent failure. A script write is immediate and fails to its caller.

Anatomy

Concept	Meaning
Node	One step. Has a type, a configuration, optional code, and input/output ports.
Connection (wire)	Passes a message from one node's output to another's input.
Message	The envelope travelling the graph: a payload plus metadata and context.
Trigger	What starts execution — a schedule, an inbound HTTP request, a tag change, a manual run.
Execution	One traversal of the graph, logged and inspectable.

The message

Every node receives a message and returns one. In scripts you see two things:

Name	Contents
payload	The primary data — what the previous node produced.
message	The full envelope: payload plus metadata, correlation identifiers and context.

def transform(payload, message):
    # payload — the data
    # message — the envelope, including where it came from
    return {"value": payload["value"] * 2}

Return the value you want the next node to receive.

Node categories

Category	Nodes
Common	Inject, Link, RBE (Filter), Switch, Timer
Realtime	Industrial I/O, Alarm
Datasource	SQL Query, MongoDB, QuestDB
Transform	Python Script, Batch Transform, Array Iterator, Join
AI	AI
Notification	Send Email, Telegram, WhatsApp, On-Call Roster
API	REST API, HTTP Response
Output	Debug

→ Node reference for every node's configuration.

Execution model

• Execution is graph-driven, not line-by-line: a node runs when a message arrives on its input.
• A node with several outbound wires fans out — every downstream branch receives the message.
• Switch is the exception: it has one output per rule, and a message leaves by the outputs whose rules matched. A branch that did not fire is skipped, and the skip cascades.
• Array Iterator splits an array into one message per item; Join collects them back.
• Batch Transform processes items of a collection in parallel.
• Timer is the only node that defers. It never blocks the graph: it hands a deadline to the runtime and returns, and the run that fires later is an independent one — which is how a repeating timer works in a graph that forbids cycles.
• Errors are logged against the execution, and the branch stops. Other branches continue.

→ Execution and triggering

Pipeline lifecycle

State	Meaning
Stopped	Not listening, not scheduled.
Running	Triggers are armed; the graph executes when they fire.
Error	Started but a node failed fatally; see the execution log.

Action	Permission
Start or stop a pipeline, or all of them	Start and stop pipelines
Execute once, now	Run pipelines
Trigger a single node	Run pipelines
See runtime state	authenticated

A pipeline is also enabled/disabled as configuration. Disabling stops it and keeps it from being started by "start all" — the right way to shelve a pipeline without deleting it.

The canvas

Control	Does	Shortcut
Undo / Redo	Step back and forward through your edits	⌘/Ctrl+Z · ⌘/Ctrl+Shift+Z
Zoom out / Zoom in		Ctrl/⌘ + scroll
The zoom percentage	Click it to reset to 100%
Zoom to fit	Frame the whole graph

New in 1.0.4 — the controls are on the canvas itself, and undo and redo grey out when there is nothing to step to, so the canvas says what is available rather than leaving you to try it.

Multi-touch pinch to zoom and two-finger pan work on a touch screen. New in 1.0.1

Editing safely

Pipelines are optimistically locked with a server-managed revision. Saving carries the revision you loaded; a concurrent save by someone else is rejected rather than silently overwriting. Two engineers editing the same pipeline get a conflict to resolve, not a lost afternoon.

Cross-pipeline communication

Mechanism	Use
Link nodes	Send messages between pipelines without wiring them into one graph.
Variables	Shared key/value state readable and writable from any pipeline or script.
Tags	The most decoupled option — one pipeline writes an internal tag, another subscribes.

→ Variables

Observability

What	Where
Per-execution log	The execution log panel (needs View execution logs)
Per-node results	The execution log panel, per node (needs View execution logs)
Live node status	Node badges on the canvas, pushed over the telemetry bus
Ad-hoc inspection	The Debug node → the Debug panel

Clearing logs requires Clear execution logs; they are diagnostic data, not the audit journal.

Permissions

Action	Permission
View pipelines	View pipelines
Create	Create pipelines
Edit	Edit pipelines
Delete	Delete pipelines
Execute once	Run pipelines
Start/stop	Start and stop pipelines
Export	Export pipelines
Import	Import pipelines
View execution logs	View execution logs
Clear execution logs	Clear execution logs

In this section

Page	Contents
Node reference	Every node type and its configuration
Execution & triggering	How and when a pipeline runs, and how to debug it
Variables	Shared state across pipelines and scripts
REST endpoints	Publishing HTTP APIs from a pipeline