Skip to main content

Understand inbound voice agents

An inbound agent handles calls placed to its assigned Wirevox phone number. A configured draft alone cannot receive calls.

Requirements for live calls​

The agent needs:

  • a successfully published version;
  • one phone number assigned in the same workspace;
  • Active runtime status;
  • an eligible paid subscription for real calls;
  • usable workspace credits and no payment lock; and
  • available concurrency under the workspace limit.

If any admission check fails, the platform does not begin a normal AI conversation. A concurrency rejection can answer with a busy message before ending.

Runtime sequence​

  1. A call reaches a managed phone number.
  2. Wirevox resolves the inbound agent assigned to that exact number.
  3. The platform loads the agent's published version, not its current draft.
  4. A call record is created and subscription, credit, payment, status, and concurrency checks run.
  5. The configured Call Answer Delay is applied, then the call is answered.
  6. The voice runtime compiles the prompt and context, starts the selected pipeline, and enables published functions.
  7. During the conversation, the runtime records events and transcript data and executes permitted actions.
  8. At the end, duration and credits are recorded and post-call processing produces the available outcome, summary, contact, and post-call actions.

Draft changes during live operation​

You can edit an Active agent without immediately changing a live call path. Autosaved changes remain in the draft until published. A call already in progress remains tied to the runtime configuration selected for that call.

After publishing, place a new call to verify the new version. Do not use a successful browser test as proof that the live number has the same version.

Call opening​

An inbound agent can use a fixed Custom Greeting. If it is empty, the model generates its opening from the prompt and runtime context. Keep a fixed greeting brief and put conditional behavior in the system prompt.

The answer delay controls when Wirevox answers the carrier call. It does not instruct the model to wait before speaking after the media session starts.

Runtime context​

Wirevox augments the authored prompt with platform context, including current date and time, caller and call details when available, configured variables, selected knowledge, voice-channel behavior, and published function definitions. Write the prompt for the resulting runtime rather than copying generated context into it.

See Variables and context for the exact context layers.

Outbound availability​

The current New Agent dialog marks Outbound Agent as Soon and prevents selection. Some internal routes and editor code exist for lead import and call initiation, but they do not constitute a supported public campaign workflow. The documentation will add outbound setup only when it is available end to end in the product.