Voice and AI overview
The model, hearing and voice behind an agent, which ones you can choose, and what you do and do not need to bring yourself.
Every agent is built from three choices, set on the Assistant tab of its page. Each has a summary card; click a card to open its settings in a panel on the right.
| Card | What it decides | Choices |
|---|---|---|
| Model | What the agent says, and most of its response delay | OpenAI and Google Gemini language models |
| Voice | How the agent sounds | Cartesia, ElevenLabs or OpenAI voices |
| Hearing | How the caller's speech becomes text | GPT-4o Mini Transcribe or three Deepgram options |
A fourth card, Cost, shows the estimated price per minute of those choices.
Built in, or bring your own?
You do not enter keys for these. Phynxa provides the language models, hearing and voices, and the dashboard has no field for an OpenAI, Deepgram, Cartesia, ElevenLabs or Gemini key. Usage is priced at the rates shown in the pickers, and the cost of every call is recorded (see Costs).
You do bring your own phone provider account (see Phone numbers) and any app you connect under Integrations or Tools.
Note. If an option shows "not available right now", or an agent shows a yellow notice such as "Deepgram speech recognition isn't available right now", Phynxa has that provider switched off for the moment. Choose a different Hearing, Voice or Language model option. A greyed-out option cannot be selected, and the server refuses it too. If you need that provider, contact us.
Language limits
Not every option works for every language. Options that cannot handle the agent's language are greyed out with the reason. See Languages.
Testing a choice
Change a card, press Save, and use Talk. Compare the speed and sound of two options by testing each with the same sentence.
Related guides
Something here does not match what you see in the dashboard? Let us know.