Cartesia Sonic for AI agents: quick voices for real conversations

What is Cartesia?

Cartesia is a voice AI company that makes speech models and a platform for voice agents that answer and make phone calls. Its text to speech family is called Sonic, and it is built on state space models, an architecture Cartesia uses to keep latency low.

Sonic voices sound natural in dozens of languages. You can choose from Cartesia's voice library or clone a voice from a short recording. Cartesia also sells a speech to text model, Ink, but SFCrew doesn't use it.

What you can do with Cartesia in SFCrew

In SFCrew, Cartesia speaks and doesn't transcribe. Everything that needs a voice works with it, and your device handles listening. The voice and calls guide shows where each control lives.

• Hear replies. Choose Speak this reply on any message, or turn on Read replies aloud in a bot's settings. Replies are spoken with sonic-3.
• Call a bot. Choose Call at the top of a chat to talk in real time. You can cut in while the bot is speaking, and the transcript stays in the chat.
• Get called. Ask "call me when the overnight list is ready" and the bot rings you. A bot may also call if you've missed the same reminder several times.
• Give each bot its own voice. The Voice next to Read replies aloud overrides your default for that bot.
• Use your own voices. The voice picker lists the voices your Cartesia account can use, including ones you cloned.
• Talk instead of typing, with your device's speech recognition. Holding the mic uses the browser's recognition in Chrome and the system's in the desktop app. On calls in the mobile app, the phone listens.

When to choose Cartesia

Choose Cartesia when you mostly listen to your bots and call them, and you want voices that sound natural and start quickly. It fits people who use Chrome, the desktop app or the mobile app, where speech recognition is already on the device.

If you need transcription from the provider too, pick one that does both. ElevenLabs has the widest choice of voices and cloning, OpenAI suits you if you already have a key, and Fish Audio adds expressive speech. To keep audio on your own hardware, run a self-hosted speech server, or use macOS voices if your server runs on a Mac. The comparison puts them side by side.

Limits

• No transcription. In a browser without on-device speech recognition, holding the mic shows an error until you switch to Chrome or the desktop app, or connect a provider that transcribes.
• Voice notes sent to a bot through a chat app such as Telegram arrive as audio files without a transcript.
• SFCrew always speaks with sonic-3. You can't pick a newer Sonic model or set emotion, speed or pronunciation from SFCrew.
• Each sentence is generated as a finished clip and played in order, rather than streamed over Cartesia's real-time connection, so a bot starts talking a little later than Sonic alone could.
• SFCrew doesn't clone voices itself. Create the clone in Cartesia first. The picker loads up to 2,000 voices, and voices beyond that can't be chosen.
• Calls and hold-to-talk are a server setting. On a self-hosted server they stay off until the admin turns them on.

How to set it up

Anyone can do this in their own settings, on SFCrew Cloud or a self-hosted server. The key belongs to your account, not the whole server.

1. Create an API key in your Cartesia account.
2. In SFCrew, open Settings → Voice, choose Cartesia as the Provider, paste the API key and choose Connect.
3. Pick a Voice and use Hear a sample to check it.
4. Optionally, open a bot's settings, expand Advanced, turn on Read replies aloud and give it its own Voice.
5. If you run your own server and want calls and hold-to-talk, set PHONE_CALLS_ENABLED=true on the API and worker.