Conditional logic

Use control-flow nodes when a workflow should branch, run paths in parallel, or merge parallel results before continuing.

What you can do here

Route a workflow down true and false paths.
Split work into two, three, or four parallel branches.
Join parallel branches back into one downstream path.
Choose how joined data is merged before the next node runs.

How to branch with Conditional

Drag Conditional from Control Flow onto the canvas.
Connect the previous node into the Conditional node.
Select the Conditional node.
Enter a JavaScript expression in Condition (JavaScript).
Connect the true and false handles to different downstream paths.

Field	Required	What it controls
Condition (JavaScript)	Yes	The expression that decides which output path runs. Use previous inputs such as `input1` in the expression.

Condition	Effect
`input1.score > 0.8`	Sends high-scoring items down the true path.
`input1.status === 'approved'`	Continues only when the previous output is approved.
`input1?.error`	Routes error payloads into a recovery path.

How to split work with Fork

Drag Fork from Control Flow onto the canvas.
Connect the previous node into the Fork node.
Select the Fork node.
Choose Number of Branches.
Connect each output handle to the branch that should run in parallel.

Field	Options	What it controls
Number of Branches	2, 3, 4	How many parallel paths the Fork node creates.

Use Fork when independent work can run at the same time, such as calling multiple agents or fetching data from multiple systems.

How to merge work with Join

Drag Join from Control Flow onto the canvas.
Connect each parallel branch into the Join node.
Select the Join node.
Choose Number of Branches to match the incoming paths.
Choose Merge Strategy.
Connect the Join output to the next node.

Field	Options	What it controls
Number of Branches	2, 3, 4	How many incoming branches the Join waits for.
Merge Strategy	Combine as Array, Merge as Object, Use First Input, Use Last Input	How branch outputs are passed to the next node.

Common patterns

Pattern	Node sequence
Error recovery	Trigger -> Agent -> Conditional -> recovery path or continue path.
Parallel research	Trigger -> Fork -> several agents -> Join -> summary agent.
First result wins	Trigger -> Fork -> multiple data sources -> Join with Use First Input.
Latest override wins	Trigger -> Fork -> enrichment steps -> Join with Use Last Input.

Connection rules

Node type	Connection behavior
Regular action nodes	Usually one input and one output.
Conditional	One input and two output paths.
Fork	One input and multiple output paths.
Join	Multiple input paths and one output.

If the builder warns that a node can only have one outgoing connection, use Fork or Conditional. If it warns that a node can only have one incoming connection, use Join.