Skip to main content

What is a Tool Step?

A Tool Step calls an external service and uses the result to guide the conversation:
  1. Execute tool — Call the external service with configured inputs
  2. Receive result — Get success/failure and any output data
  3. Map outputs — Store results in variables for later use
  4. Evaluate conditions — Navigate based on tool success or failure

Tool Step Structure


Input Mapping

Input mapping passes values to the tool. Values can be:
  • Static text — Fixed values
  • Variables — Contact and live variables from the conversation (type / to see available variables)

Tool Results

Tools return results that can be used in conditions and subsequent steps. The assistant has access to tool outputs and can incorporate them naturally into the conversation:

Conditions

Tool steps use boolean conditions based on success or failure:
A tool condition can also end the call outright: its Next step… dropdown offers End Call and End Call (Strict) alongside the steps and the End Flow outcomes. See Where a condition can send the call.

Classifiers on a tool step

A tool step can carry up to three Classifiers — background AI extractions that run alongside the tool and write their findings into variables without changing where the call goes. On a tool step they finish before the step moves on. Add them from the step’s menu; see Classifiers.

Runaway loops

A chain of tool steps that routes back on itself can, in the wrong circumstances, keep running with no caller turn in between. Voxworks stops this on its own: after a long run of consecutive tool steps with nobody speaking, the assistant apologises and ends the call rather than spinning. If you see this on a call, look for a tool condition that routes back into an earlier tool step with no voice step in the loop.

Interrupts

Tool steps cannot be interrupted. While a tool is executing, this ensures tool operations complete reliably without being disrupted mid-execution. Use the Script to provide a “thinking” message like “Let me check that for you…” which will happen whilst the tool executes, so it is OK to keep it brief.

Best Practices

  1. Handle failures gracefully — Always have a fallback condition
  2. Validate inputs first — Use a Code Step to validate data before tool calls
  3. Inform the user — Use the script property to set expectations (“Let me check that for you…”, “Let me see…” or “Ok…”)
  4. Use tool outputs — The assistant can reference tool results in responses
  5. Test integrations — Verify tools work with your specific configuration

Next Steps