Template Expressions

Template expressions allow nodes in an agent pipeline to reference data from the input or from other nodes' outputs. They use the {{...}} syntax and are resolved at runtime just before each node executes.

Syntax

{{source.path.to.value}}

All expressions are wrapped in double curly braces. The first segment identifies the data source, and subsequent segments navigate into the data structure using dot notation.

Data Sources

Input Data

Reference the agent's invocation input:

{{input.message}}          // Top-level property
{{input.user.name}}        // Nested property
{{input.items[0].title}}   // Array access

Node Outputs

Reference the output of any upstream node by its ID:

{{nodes.llm_1.output}}          // Full output of node llm_1
{{nodes.tool_1.output.data}}    // Nested property
{{nodes.transform_1.output}}    // Transform result

Loop Context

Inside a Loop node's body, access iteration data:

{{loop.item}}         // Current item being iterated
{{loop.index}}        // Zero-based index
{{loop.item.name}}    // Property of the current item

Examples

Simple String Interpolation

You asked: {{input.message}}

Resolves to: You asked: What is the weather?

Composing Multi-Node Data

Based on the search results ({{nodes.tool_1.output.count}} found),
here is a summary: {{nodes.llm_1.output}}

JSON Construction in Transform Nodes

const query = "{{input.query}}";
const results = {{nodes.tool_1.output}};
return {
  query,
  resultCount: results.length,
  summary: results.map(r => r.title).join(", ")
};

Type Handling

Template expressions resolve to the actual JavaScript type of the referenced value:

When an expression is used inside a string (like a prompt template), non-string values are automatically serialized to JSON.

When an expression is the entire value of a field (like a Transform node's input), the original type is preserved.

Error Handling

If a template expression references a node that:

• Does not exist — Pipeline validation will catch this before execution
• Has not executed yet — Topological sorting ensures this cannot happen for valid pipelines
• Returned an error — The downstream node will receive null and the execution log records the failure

To handle optional data, use JavaScript in Transform nodes:

const data = {{nodes.tool_1.output}};
const name = data?.name || "Unknown";
return { name };

Best Practices

1. Keep expressions simple — Use Transform nodes for complex data manipulation
2. Validate your schema — Define Input node schemas to catch bad data early
3. Use meaningful node IDs — search_api is clearer than tool_1 in expressions
4. Test incrementally — Use the agent test panel to verify expressions resolve correctly