Workflows

A workflow is a graph the engine runs in topological order, piping each node's output into its downstream neighbours.

The model

A workflow is { nodes, edges }. Exactly one trigger node starts it; the engine walks the graph in dependency order and pipes each node's output to the next. The graph must be a DAG.

Node types

TypeDoes
trigger.*Entry point: manual, scheduler (cron), webhook, channel, or event.
agentRuns an agent (multi-step, uses its own tools).
skill / plugin.actionCalls one registered action/skill.
conditionBranches true/false — the untaken path is pruned.
set_variableWrites ${vars.*} for all later nodes.
human.handoffPauses the run for a person; resumes on decision.
endTerminal; its input is the final result.

See Nodes & data flow for the full interpolation model, and the dedicated pages on branching, error paths, variables, and human-in-the-loop.