Skip to main content
Live Voice is a two-way voice conversation. Your Agent can answer aloud, perform tasks with its enabled tools, and keep transcripts in the conversation history. Ordinary Dictation instead turns speech into text input.
Realtime Voice access and automatic hang-up settings

Realtime Voice access and automatic hang-up settings.

​
Set up voice access

  1. Open Settings → Agents → Realtime Voice.
  2. Choose Enconvo Cloud or your own provider under GPT Live → Voice Access, according to the options in your build. For access with your own key, select the OpenAI provider profile whose API key should be used.
  3. Confirm that Enconvo has Microphone access in macOS and select working microphone and speaker devices.
  4. Check the idle timeout, which controls when an unused call hangs up.
Realtime Voice access is configured separately from an Agent’s ordinary text model. A text-model subscription connection does not by itself establish GPT Live API access.

​
Start and end a call

  1. Open the Agent and conversation you want to use.
  2. Click the voice waveform / Live Talk control. Wait until the connection indicator shows that the microphone is ready.
  3. Speak a complete request. For a tool task, name the desired result and provide enough context for the Agent.
  4. Expand the working details to inspect tool progress or results.
  5. Use the microphone control to mute, or the hang-up control to end the call.
Spoken turns are saved in regular chat history. Reopening that conversation lets you review the dialogue and work details. You can also use an Avatar as the visible speaker.

​
If a call does not work

  • It cannot connect: check the selected voice access method and OpenAI provider profile, then read the connection error.
  • The microphone is unavailable: finish an active Dictation or microphone recording session first, and check macOS Microphone access.
  • The Agent cannot perform an action: review that Agent’s tools and permissions. Voice does not add tools the Agent lacks.
  • The call ends while idle: review the idle timeout in Realtime Voice settings.
See Agents for tool configuration and Speech Recognition for transcription providers.