Action Inputs
How data moves between actions, how to choose an action's inputs, and the data types SPARK checks along the way.
A workflow is a chain, so most actions work on data produced by the actions before them. This page covers how that data flows, how to pick an action's inputs explicitly, and the lightweight type system SPARK uses to catch mismatched connections. For the building blocks themselves, see Build a workflow.
Data flows along the connections
By default, data follows the arrows you draw on the canvas: each action reads the output of the action(s) connected into it, and passes its own output on to the next. For a simple linear chain you don't configure anything, the connections carry the data.
Choose an action's inputs
When an action should read from specific earlier steps rather than just its immediate connection, set them in the side panel's Data Sources section. It is folded until you click its heading; the count and any error stay on the heading while it is folded.
- The description reads Select which actions provide data to this action. Leave empty to use canvas connections.
- Open the Select actions... picker and choose one or more sources. Only actions that come before this one on the same branch are offered.
- A badge shows how many sources are selected against the maximum, for example 2 / ∞. Some actions require a minimum and warn until it's met (N data source(s) required, …).
Each selected source shows as a row under the picker, with a remove control (Remove source). The picker names each earlier step as it was added, for example Load records. An action that needs no data source at all, because it reads from a data object or a file itself, has no Data Sources section.
A few actions instead take a single named source through a Select source action... picker, for steps that combine two inputs (such as a reconcile or merge).
Data types SPARK checks
Every action has an input type and an output type. They show as an icon pair on the action's entry in the Configure list, and beside its group badge at the head of its panel. The types are:
| Type | Meaning |
|---|---|
| Table | Rows and columns, such as parsed CSV records |
| File | A file reference |
| Text | Free text |
| Any | Untyped; connects to anything |
These types keep connections sensible:
-
When you pick Data Sources, sources whose output type can't feed this action are hidden.
-
If a connection's types don't line up, the edge turns into a dashed line with a Type mismatch marker; hover it for a message like Ask Agent outputs Text but Save Record expects Table. Clicking actions in turn still connects them when the types clash, so the mismatch is visible as soon as you make it. It's a warning while you edit, not a hard stop, and picking a matching Data Sources entry clears it.
-
Saving a workflow with a real type mismatch is rejected, so the check is enforced when you save.
Feed a step the previous output
The Ask Agent action has a Use previous action data toggle, on by default, that passes the prior step's output into the agent's prompt, so the agent reasons over what the workflow just produced. With Structured output on, its answer becomes a Table that steps such as Save Record read directly; without it, the step outputs Text. See Ask an agent.
Read what a step produced
A step's output is not only for the next step. In any run, click the step to open its Output: a line reading, for example, 10 of 52 records, with Export and View all at its right, above a table of the first rows. View all pages through every row. Large outputs are read a page at a time, so even a step of many thousands of rows opens. See Executions.