Structured Procedures

A Structured Procedure compiles authored steps into a deterministic runtime graph. It is suited to regulated checklists, required data collection, exact notices, and controlled Tool execution.

Build the sequence#

Set the name and trigger, then add steps. Ask generates a question from an instruction and waits for the next customer message. Tell generates a response from an instruction but does not automatically wait. Say emits the exact authored message. Tool calls one selected webhook Tool. If chooses one branch. End either closes the conversation or returns. Sub-procedure inlines another Procedure and later returns to the caller.

Steps are executed in authored order. An If branch rejoins the following step unless an End changes control flow. Sub-procedures are expanded when readiness is checked; cycles and missing references are rejected.

Waiting and returning#

Ask is the normal pause boundary: Operade stores the current internal node and resumes it when the customer replies. Tell and Say continue automatically, so long automatic sequences should be short. An End step with End conversation enabled closes the conversation. With the option off, the current Procedure returns to its caller, Agent, or Workflow.

The runtime allows no more than 20 consecutive automatic Procedure steps before a wait, and expanded authored execution is capped at 100 steps. These limits prevent silent loops and excessively large compiled graphs.

Validate before publishing#

Draft autosave accepts incomplete work and displays field-level readiness errors. Publish and Test chat require instructions, messages, Tool IDs, branch expressions, and Sub-procedure references to be valid. Missing Tools are rechecked at runtime as well.

See the complete step reference and runtime troubleshooting.