Workflow builder
Choose a few templates for the current page or text, then combine their analyses into one useful answer.
InData
Optional. If empty, this step uses the widget-selected model before any template-recommended model.
Runs as a second AI call: the template answers first, then this rewrites its answer. Leave it empty and the step costs one call.
Runs that workflow and uses its final answer as this step's output. It is a live reference, not a copy: improving that workflow improves this one. If it is deleted or made private, this step says so in its output instead of failing the run.
Fetches page content from the URL. The URL is this step's input, so InData wiring is not used.
This text is this step's input, so InData wiring is not used. The body is used literally — {$Step name} references are not resolved. Long references reach the model whole and can overflow small or on-device models.
The reader's own profile, limited to the parts you tick. Nothing is read unless a box is ticked, and the reader can narrow it further when they run the workflow. This step is its own input, so InData wiring is not used; later steps can read it as {$Step name}.
Optional. Ask this step for named fields instead of loose prose, so a later
step can read {$Step name.field} and a condition can compare one.
The step still produces readable text, so {$Step name} keeps working.
No fields. This step answers in prose.
Optional. Runs this step once for every item in a list an earlier step promised,
with {$item}, {$index} and {$count} filled in
per run. Each run is its own AI call, so a list of twelve is twelve calls; long
lists are capped.
Optional. A skipped step costs no AI call and its output does not reach the final answer. Reading a field needs that step to promise one under “Answers with”; leave the field empty to search its whole answer.
If enabled, this step will not start with the current group. It starts only after every step in the current batch has finished.