Skip to content

Building Flows

Build a Flow in Canvas → Design. Start with the smallest topology that expresses the task, then add consequence-bearing blocks only when needed.

A practical build sequence

  1. Add a source or another block that can begin with available rows.
  2. Configure each block before connecting a long chain.
  3. Connect outputs to downstream inputs in the intended order.
  4. Review validation findings and current workspace mounts.
  5. Use Inspect to examine the selected block and nearby context.
  6. Save when you want a durable identity and revision.
  7. Run only when you intend to create execution evidence and effects.

The current Flow contract has one logical rows channel. Connections serialize as upstream dependencies; the drawing does not create arbitrary named edge payloads.

Configuration and consequences

Deterministic row transforms are a good default for filtering, mapping, expanding, aggregating, ranking, or routing data. Model, Script, network enrich, and Memory Sink blocks introduce provider, execution, remote, or persistence boundaries. Their configuration proves intent, not that the consequence occurred.

Use Block Reference for supported configuration and runtime notes.

Visual and code views

Visual and JSON/YAML code views represent the same draft. Code view expects a job object containing spec.blocks[]. Invalid code remains an authoring error; it is not partially executed. Use visual view for topology and the expanded Script editor for substantial native code.