Data flows through a workflow along its edges. Each node produces an output, and any node downstream can read that output by referencing it with a variable tag. In the builder these tags appear as friendly pills, so a reference reads as “input from Summarize” instead of raw bracket syntax.
The two tag formats
Every reference resolves to one of two canonical forms. You rarely type these by hand, since the “/” picker inserts them for you, but it helps to know what a pill stands for.
[Input from node-1] the entire output of node-1
[[fieldName] from node-1] a single named field from node-1Use [Input from node-1] when you want the whole result of an earlier step. Use [[fieldName] from node-1] when that step produces several named fields and you want just one of them.
Inserting a reference
Click into any field that accepts variables and type /. A menu lists the upstream nodes you can pull from. Choose one and the correct pill is inserted at your cursor. You can mix pills and normal text freely, for example:
Write a product description for [[title] from ai-1] in a friendly tone.Structured multi-output from the AI Prompt node
By default the AI Prompt node returns one block of text, which you read with [Input from node-id]. When you need the AI to return several distinct fields, turn on structured output on that node and define an output schema: a list of fields, each with a name, a type (string, number, boolean, array, or object), and a short description.
With structured output on, the node returns a named field for each schema entry, and you reference each one directly:
[[title] from ai-1]
[[excerpt] from ai-1]
[[body] from ai-1]This is the preferred way to split one AI result into several fields, for example to fill a post’s title, excerpt, and content from a single AI Prompt node. You do not need a separate Parser or Extract node in between. See the AI Model Node for how to define the schema.
Extracting named fields from any text
When the text you want to split did not come from a structured AI node, for example a form submission or a scraped page, use the AI Extract Information node. You define extraction fields, each with a name, a description, and a flag for whether it is a single value or a list. The node then produces one named output per field, which you reference the same way:
[[email] from extract-1]
[[topics] from extract-1]Named fields from triggers
Form, webhook, and RSS triggers expose their fields by name. A form trigger, for instance, gives you each submitted field by its label, so you can reference [[Email] from trigger-1] or [[Message] from trigger-1] directly. A manual trigger produces a single text payload you read with [Input from trigger-1].
A note on plain text output
If structured output is off, a plain AI Prompt node returns prose, not named fields. A [[field] from node-id] tag against it resolves only if that node’s prompt explicitly tells the model to return strict JSON with that exact key. When in doubt, turn on structured output or use an Extract Information node, both of which produce reliable named fields.