Skip to main content

DSL Semantics

This page explains how the platform interprets the DSL at runtime.

1. Journey Ordering​

The journey array is the default execution path. A step runs, then the next step runs unless a branch or jump changes the path.

2. Continue vs Targets​

Some steps return outcomes that point to targets:

  • continue means go to the next step in the journey
  • done ends the conversation successfully
  • escalate hands off to a human
  • goto jumps to a named step

3. Interactive vs Instant Steps​

Interactive steps wait for user input:

  • collect, choice, confirm, guide

Instant steps execute immediately:

  • action, investigate, set, if

4. Naming And Goto​

A step can include name, which creates a slug used by goto.

- collect:
    name: Ask Again
    fields: [order_id]

- goto: ask-again

5. Variables And Data Flow​

  • collect stores values in variables
  • action results are stored in variables using as:
  • set writes computed values into variables

Use {{variables.*}} to interpolate values in messages. Use $variables.* in expressions.

6. Triggers And Routing​

The router matches a user message to a capability using:

  • trigger.goal
  • trigger.examples
  • optional trigger.priority
  • optional trigger.requires

7. Knowledge Usage​

A capability may attach knowledge collections. When knowledge is available, the system retrieves relevant chunks and grounds responses.

8. Fallback Behavior​

If a journey cannot continue, the capability fallback is used. If not set, the platform escalates to the default queue.

9. Auditable Execution​

The platform records routing, step transitions, and action calls in audit logs. This makes execution traceable without granting the LLM control over decisions.