JSON Formatter

The JSON Formatter step gives you exact control over your JSON payloads in Agent Studio. Build them visually or hand‑craft them in a raw editor, then pass the result to downstream steps.

TL;DR

  • Two modes: Visual Builder and Raw JSON Editor.
  • Use cases: Shape final payloads, normalize LLM output, or build arrays for Loops.
  • Where: Steps → Data Formatter → JSON Formatter.
  • Output: A valid JSON value (object/array/primitive) resolved at runtime.

When to use the JSON Formatter

Use the step when you need to:
  • Guarantee a specific schema for downstream systems (APIs, db writes, exports).
  • Normalize/clean LLM output into predictable keys and types.
  • Assemble arrays of items for a Loop (e.g., iterate over items[]).
  • Rename keys or remap fields without touching previous steps.
If you only need to pass a single variable through, you can skip this step; but for anything schema‑related, the JSON Formatter is the safest place to lock it in.

Modes

Visual Builder (no‑code)

An interactive canvas for assembling JSON with key/value pairs and nested structures. Type or paste variable tokens using the <stepResult value="..."/> syntax (see below). Highlights
  • Add objects and arrays with one click.
  • Insert variables without syntax mistakes.
  • Inline validation + live Preview of the resulting JSON.
Great for: quick mapping, small–medium payloads, avoiding syntax errors.

Raw JSON Editor (power users)

Write or paste any valid JSON. The editor validates the syntax and shows the structured preview. Highlights
  • Full control over nested objects/arrays.
  • Easy copy/paste from specs or other tools.
  • Supports the same token syntax (<stepResult value="..."/>).
Note: The editor validates JSON (not JavaScript). Use double quotes for keys/strings and no trailing commas.

Variable tokens (syntax)

Variables are inserted as self‑closing tokens. For example, referencing a previous step result named AI Model inserts:
<stepResult value="AI Model"/>
  • Type tokens manually using the syntax below. Copy/paste to avoid typos.
  • In Visual Builder, type or paste tokens into the value field. The builder handles escaping and quoting for you.
  • In Raw JSON Editor, place tokens inside JSON strings so the JSON remains valid.
At runtime, tokens resolve to their actual values before the JSON is sent to downstream steps.

Quick start

  1. In Agent Studio, add JSON Formatter from Data Formatter.
  2. Choose Visual Builder or Raw JSON Editor.
  3. Type the token syntax (e.g., <stepResult value="AI Model"/>).
  4. (Optional) Add arrays/objects and nest as needed.
  5. Connect the JSON Formatter’s output to an Output or a Loop step.

Validation & preview

  • The editor validates the JSON structure on the fly.
  • The Preview panel reflects the structure with tokens in place; during execution, tokens resolve to values.
  • Any errors appear inline; fix and continue without losing your work.

Best practices

  • Freeze your schema before integrating with external systems.
  • Name keys for consumers, not creators—keep a consistent case (snake_case or camelCase).
  • Keep arrays homogenous (same keys/types per item) to make Loop logic simple.
  • Use the token syntax consistently; copy/paste to avoid typos.

Troubleshooting

“Invalid JSON” in Raw Editor
  • Tokens must be inside quoted strings (e.g., "<stepResult value=\"AI Model\"/>").
  • Check for trailing commas, single quotes, or missing quotes on keys.
Token didn’t resolve
  • The referenced step name must match exactly (e.g., AI Model). Re-type the token exactly or copy/paste from the source step name to be sure.
Loop doesn’t iterate
  • Confirm the JSON Formatter outputs a top‑level array (e.g., [ {...}, {...} ]).
Unexpected nulls
  • The source variable may be missing at runtime. Provide a default value earlier or change the mapping.