Skip to main content
Fanout takes a JSON array as input and runs the downstream node once per element, in parallel. When all parallel runs complete, the results merge back into a JSON array and flow onward. The fan-in happens automatically: the node immediately following the fanned-out node collects all results into an array, preserving order. No manual collection step needed. Results that are JSON join the array as JSON, and numbers like 5 or 1.5 join as numbers. Everything else joins as text, exactly as the node produced it, so a version like 1.10 or a long ID stays intact. To do arithmetic on values like 12.50 in a JSON Query, convert them first: sum(map(&to_number(@), @)) in JMESPath. Example: You have a list of 50 customer questions. Fanout splits the array, the downstream Prompt node processes each question independently and in parallel, and the results recombine into a single array of answers.

​
Limits

A single fanout handles up to 1,000 items, and a workflow run creates up to 5,000 node runs in total – the second limit catches fanouts that multiply, since two fanouts feeding the same node produce one run per combination. Whichever limit trips is checked before that node expands, so none of its runs are created or billed – work the run completed earlier still is.
Need to process more? Chunk the array upstream and run the workflow once per chunk – each run gets its own budget.