Data Flow in Workflows
Understanding how data moves through a Flux workflow is key to building effective automations.
Step Outputs
Each step produces an output when it completes. For agent steps, the output is the final response from the LLM after it has finished using tools and reasoning. For action steps, the output is the result of the deterministic operation (e.g., email send confirmation).
Accumulated Data
Flux maintains an accumulated data object throughout the workflow run. After each step completes, its output is added to this object under a key derived from the step ID. All subsequent steps can access this data.
Output Schemas
You can define structured output schemas on agent steps to ensure the agent produces data in a specific format. This is especially useful when downstream action steps need to parse the output — for example, extracting a list of email addresses or a set of numerical metrics.
Config References
Action steps use the config.batch_input field to reference data from previous steps using dot notation. For example, step_abc123_output.emails resolves to the emails field from the output of the step with ID abc123. This makes it easy to wire up multi-step pipelines without custom code.