Skip to main content
A reference chip in a prompt with its preview showing the value and Go to node, Replace and Remove
Sometimes drawing an edge isn’t what you want. You want to drop another node’s output value directly into the middle of a prompt or template string. References do this without adding an edge. For an introduction to inline references and how they sit alongside edges and published parameters, see Connecting Nodes.

Template syntax

References use the syntax {{nodeId.parameterKey}} to embed the value of another node’s output parameter. When the node runs, Runchat replaces each template with the current value of the referenced parameter. You don’t usually type these directly. Use the @ picker in any text input.

Inserting a reference

In any text input, type @ to open the reference picker. It has three sections:
  • Connected to this node: outputs from nodes already wired into this one
  • Elsewhere on the canvas: outputs from every other node
  • Secrets: your saved secrets, for passing API keys without pasting them into the prompt
Keep typing to filter the list. Use the arrow keys and Enter, or click, to insert a chip representing the current output value. To delete a chip, press backspace next to it. Hover or click a chip to preview the value it points to. Go to node selects and zooms to the source node, Replace points the chip at a different output, and Remove deletes it. If the source node has been deleted, the chip tells you so. Replace or remove it before running.

Run order and propagation

References behave differently from edges in two important ways:
  1. Changing a referenced value does not automatically expire downstream nodes. With an edge, when an upstream node finishes, downstream nodes are marked as ready to re-run. References do not propagate this signal. The downstream node will only pick up the new value the next time it runs.
  2. When you do run the workflow, references still affect order. Runchat builds the run order from both edges and references, so a node that references another will always run after it during a multi-node run.
If you want a node to update automatically when an upstream value changes, draw an edge instead.

Next steps

  • Connecting Nodes: edges, references, and published parameters compared
  • Data Types: what flows through a reference and how it’s cast
  • Routing: conditionally enable parts of a workflow