Templates and expressions
Most text settings on a node take templates: prompts, URLs, request bodies and headers, Set field values, response templates, file names. Everything between {{ and }} is replaced when the node runs.
The body of a {{ }} is one of two things:
- A path: letters, digits,
_,-and dots, such as{{user_prompt}}or{{results.fetch.items.0.title}}. A number segment picks an array item. - An expression: anything else, such as
{{ answer.trim() }}or{{ $if(score > 90, 'high', 'low') }}. Expressions run in a sandbox with a time and memory limit (code execution).
A string renders as it is, a number or true/false as text, and an object or array as JSON.
What a template can read
A node's template sees the steps connected straight into it, the run's input, and every step that has already finished, but each through its own form:
| Form | What it reads | Example |
|---|---|---|
{{field}} | A field of a step connected straight into this one (a direct parent). | {{answer}} |
{{nodeId.field}} | The output of a direct parent, by its node id. | {{fetch.status}} |
{{ $json.nodeId.field }}, {{ $input.… }} | The same data as the two forms above, in an expression. | {{ $json.fetch.status }} |
{{parent_outputs.nodeId.field}} | A direct parent's output, by its node id. | {{parent_outputs.fetch.status}} |
{{results.nodeId.field}} | The output of any step that has already finished in this run. | {{results.fetch.body.title}} |
{{ $node('nodeId').output.field }} | The same as results, in an expression. | {{ $node('fetch').output.status }} |
{{user_prompt}}, {{system_prompt}}, {{session_id}}, {{webhook.field}}, {{trigger_type}} | The run's input and trigger payload, in every node. | {{webhook.customer.id}} |
{{vars.name}}, {{ $vars.name }} | Workflow variables, in every node. | {{vars.API_URL}} |
{{workflow.id}}, {{workflow.name}}, {{execution.id}}, {{execution.customData.key}} | The workflow and this run. $workflow and $execution work in expressions. | {{execution.id}} |
The node id is the id in the workflow JSON, not the label on the canvas; nodes added in the editor get ids such as http_request-k3j9x2. A path body takes such an id as it is: {{results.http_request-k3j9x2.body}}. In an expression, a - is a minus sign, so write results['http_request-k3j9x2'].body or $node('http_request-k3j9x2').output.body. A direct parent's field with the same name as a run input (a parent that outputs user_prompt, say) is the one {{user_prompt}} reads.
Direct parents and earlier steps
{{field}} and {{nodeId.field}} follow the item from step to step: a node sees what its direct parents produced, not what came before them. Two steps away, they render empty.
fetch → summarize → reportIn report, {{summarize.text}} works, but {{fetch.body}} renders empty because fetch is not a direct parent of report. Use {{results.fetch.body}} (or {{ $node('fetch').output.body }}) to read any step that has already finished, wherever it is in the graph.
Steps that pass their input on keep the parent's data in reach: after a No-op, Wait, Filter, IF, Switch or Chat Intent Router step, or a disabled node, for example, {{fetch.body}} still works in the next node. parent_outputs holds the direct parents only, so {{parent_outputs.fetch.body}} does not see through them.
A step that has not run yet (it comes later, or its branch was skipped) is not in results, unless its output is pinned: reading it renders empty.
If a step checks a value with content rules, read the value that step passes on ({{field}} in the node after it). {{results.<id>.field}} and $node('<id>') of a step before the check read the value from before the check.
Loops
The nodes connected straight after a Loop / ForEach node run once per item. Only those repeat: a node after them runs once, after the loop. Inside a pass, {{item}} (or the name set in Item variable) is the current item and {{_loop_index}} its position, from 0.
After the loop, a loop body node's result is its last pass: {{results.<bodyId>.field}}, and the nodes after it, see that pass only. To use every pass, read the Loop node's output: {{results.<loopId>.items}} is a list with one entry per pass, each with index, item and outputs.<bodyId>; {{results.<loopId>.count}} is the number of passes.
{{ results.loop.items.map(pass => pass.outputs.summarize.text).join('\n') }}When a value is missing
A template never fails a node for a missing value: a path that does not resolve renders as an empty string, with no warning. So does an expression that throws or runs out of time, such as {{ $node('fetch').output.body.title }} before fetch has run. Only a failure of the sandbox itself fails the node (code execution).
A path only walks objects and arrays: on a text, {{answer.length}} renders empty. Write it as an expression instead, such as {{ answer.trim().length }}.
When a template renders empty, check, in this order:
- The node id: open the step and compare its id with the one in the template.
- Whether the step is a direct parent. If not, use
{{results.<nodeId>.field}}. - Whether the step ran before this one in the last run (the run's trace shows each step's output).
- The field path: the step's Output tab in the inspector shows the exact shape.
The editor's preview under each template field renders it as a run would, with the last run's data: the node's recorded input, and every step's output as results.
Expressions
Expressions are JavaScript with these names available:
| Name | Value |
|---|---|
$json, $input | The node's template data: its direct parents' fields, the run input, results, vars, workflow, execution. |
$node('nodeId').output | The output of any step that has already finished (the same as results.nodeId). |
$vars | Workflow variables. |
$workflow, $execution | id and name; id and customData. |
$now, $today | The current time, and today at midnight, as Dates. |
$if(condition, a, b), $ifEmpty(value, fallback) | Small helpers. |
$jmespath(value, 'path') | A JMESPath-style lookup. |
$env | A small, curated set of environment values (never secrets). |
Every top-level field of the template data whose name is a valid JavaScript name is also a name of its own, so {{ answer.toUpperCase() }} and {{ results.fetch.status === 200 ? 'ok' : 'retry' }} both work.
Related
- Core nodes and the node reference: which fields each node has.
- Code execution: the sandbox, its limits and the Code node.
- Data labels: a template read carries the labels of the step it reads.
- Common issues: a template two steps away renders empty.