Skip to main content

Path to production

This tutorial takes a voice AI Agent from zero to live traffic. Each step links to the page with the details.

Steps​

  1. Create the AI Agent. Create an AI Agent and add its knowledge and tools. You use the same AI Agent for voice, chat and email. See Build your first AI Agent.

  2. Write voice guidelines. Tell the AI Agent to use short spoken sentences and to confirm important data. Remove lists and formats that do not work in audio. See Voice guidelines.

  3. Set the voice settings. In Settings → Languages → Speech configurations, open the row of the primary language. Select the speech-to-text and text-to-speech provider, the model and the voice. Open another language row to give that language its own voice. For ElevenLabs, use Test voice to hear the voice. See Voice settings.

  4. Set the Voice model. In the Voice channel of the model strategy, select one language model. Voice uses no fallback. Use Low or Medium reasoning effort to keep replies fast. See Model strategy.

  5. Add speech models (optional). To use your own provider account, add its key in Settings → Developers → Speech models. Then select it in the voice settings. See Speech models.

  6. Connect a voice channel. For phone calls, go to Connections, connect your Twilio or Telnyx account, and select the phone number. For Twilio, use the credentials of the region that matches your Moveo account. See Twilio Voice or Telnyx Voice. For calls from your website, connect Web Voice to the web widget. Web Voice needs no phone number or telephony account. Connect both to take calls on the phone and on the website.

  7. Route calls with a rule. Create a rule in the environment that sends calls from the integration to the AI Agent. Use a staging environment that routes to the draft. See Rules and Staging and live environments.

  8. Test. Make a Web Voice call and a phone call. Listen to the greeting, the turns and the end of the call. Then run simulations with the Voice channel type. See Test voice agents.

  9. Publish a version. Click Publish on the AI Agent and describe your changes. Then select the new version in the rule of the live environment. See Versions.

  10. Monitor. Read the call metrics in analytics. Open the AI Agent logs to read transcripts and to play call recordings. To find slow turns, see Latency.

  11. Roll out. Move the phone number of your live line to the integration, or forward calls to it. For outbound calls, start with a campaign to a small list. See Outbound calls.

Before you go live​

  • The AI Agent ends calls with the Resolve action, with a 1 to 2 second Pause before it.
  • The Handover action has a destination, and the workflow does not run it on Web Voice calls.
  • Store call recordings is on or off in the integration, as your policy requires. See Call recording.
  • Simulations with the Voice channel type pass on the version that you publish.
Last updated on