Voice agents
A Wirevox voice agent answers a telephone or browser call, understands the caller, reasons with the configured model and context, speaks through the selected voice, and can use attached functions.
The creation dialog currently supports Inbound voice agents. Outbound Agent is visible as Soon and is not selectable. Do not build a production campaign around an internal or incomplete outbound path.
The complete voice-agent stack
| Layer | Responsibility |
|---|---|
| System prompt | Role, outcome, conversation policy, grounding, action rules, escalation, and boundaries. |
| Architecture and model | Runs the conversation through a Modular or Realtime voice pipeline. |
| Speech configuration | Controls language, recognition, voice, delivery, timing, interruptions, and silence. |
| Runtime context | Adds date, time, caller context, configured variables, knowledge results, and available functions. |
| Functions | Perform approved actions such as collecting data, scheduling, messaging, or transferring. |
| Published version | Immutable configuration used for real calls. |
| Phone routing and status | Connects an assigned number to an Active inbound agent. |
| Operations | Produces call records, transcripts, outcomes, analytics, and billed usage. |
In this topic
- Understand inbound voice agents
- Configure a voice agent
- Add knowledge, variables, and functions
- Test and optimize voice behavior
- Publish, connect, and operate
- Troubleshoot voice agents
Recommended build order
- Define one primary call outcome and the escalation path.
- Create an inbound agent from the closest template or Blank.
- Write the prompt and stable variables.
- Select the architecture, model, language, and voice.
- Attach only the knowledge and functions the role needs.
- Test ordinary, ambiguous, interrupted, silent, and failure paths.
- Review the draft cost, publish a named version, and assign a number.
- Start the agent, place a real call, and inspect the resulting log.
Important lifecycle rule
Browser tests use the saved draft. Real calls use the published version. Saving, publishing, assigning a number, and starting the agent are separate operations.
For the state model, see Agent lifecycle. For a guided first setup, see Create your first voice agent.