Book a build call
Back to Help
Guide

Give it a voice that speaks first

4 min read

A NetShow agent is built to talk. It has a voice of its own, it can open the conversation with a first line you write, and a visitor can answer out loud or in writing. This guide shows where each of those is set.

Where its voice is chosen

Open the agent editor from your agents list (the wand button titled Unified Editor), then choose the second row of tabs, Channels.

Open the Realtime tab. Under Realtime Model & Voice there is a menu called Realtime Voice. That is the voice your agent speaks with in a spoken conversation. Pick one and press Save.

Next to it you will also see a Voice tab with its own Voice Selection menus. The agent's spoken voice is read from Realtime Voice, so that is the one to change. When a voice sounds wrong, check the Realtime tab first.

Hear it before a visitor does

At the top of the editor, the Talk button opens a spoken conversation with your agent. Use it after every voice change: one sentence tells you more than any description of a voice can. Chat opens the same agent as a typed conversation.

The first thing it says

An agent that waits in silence feels like an empty shop. Give it an opening line.

Go back to the first row, Who Is Your Agent, and open Personality. Under Welcome Message:

  1. Turn on Enable Welcome Message. The page describes it exactly: "Automatically send a greeting when a user starts a session."
  2. Write the line in the Welcome Message box.
  3. Press Save.

A good first line is short, says who is talking and asks one easy question:

Hi, I'm Maya from Riverside Dental. Are you booking a visit or asking about one?

Keep it to one or two sentences. A visitor decides in seconds whether to answer, and a long opening gets talked over.

The same tab has a Script section (Enable Script) for a scripted opening flow. Most agents do not need it; a clear Welcome Message and good Knowledge Text go further.

The visitor's microphone is theirs

Speaking to your agent is always the visitor's choice. A browser never lets a page hear anyone without asking: the visitor is asked for microphone permission first, and nothing they say is heard until they allow it. A visitor who says no, or who has no microphone or speaker, can type to the agent instead.

Phone calls are a separate page

The editor sets how the agent sounds. Giving it a phone number happens on Voice & Phone Setup, at /dashboard/voice-setup. The old addresses /dashboard/voice and /setup bring you to the same page.

At the top are three buttons: Manage Agents, Manage Numbers and Test Call. Below them the page counts your Phone Agents, Available Numbers and Assigned Numbers, then offers a Setup Wizard:

  1. Step 1: Select Agent, the phone agent that should answer.
  2. Step 2: Assign Number, one of your available numbers.
  3. Assign, then Test Call to hear it answer.

If you have no phone agent or no number yet, the wizard says so in one line and the page offers Create Phone Agent and Add Number instead.

What this guide does not cover

When the voice and the first line both sound right, publish it.

Still stuck?

The longer documentation covers models, voice, knowledge, and workflows in depth.

Open the documentation