# Voice and calls

Talk to your bots instead of typing, have them read replies aloud, or call them.

## Set up a voice provider

Open **Settings → Voice**, pick a **Provider** and add its **API key**:

- **ElevenLabs**, **OpenAI** and **Fish Audio** can speak and transcribe.
- **Cartesia** speaks only.
- **Self-hosted** uses your own OpenAI-compatible speech server, such as Kokoro or Whisper. Enter its
  **Server URL**, and set its **Speech model** and **Transcription model** under **Models**.<!-- self-hosted -->
- **macOS** uses the voices installed on the Mac running your server, with no key. It appears only
  when the server runs on a Mac.<!-- /self-hosted -->

Then choose **Change** next to **Voice** to pick a voice, and use **Hear a sample** to check it.

Voice settings apply to every [space](/docs/spaces/). A space can customize them in its settings.

## Talk instead of typing

Hold the microphone in the composer and speak; release to send. Slide up to lock recording so you
don't have to keep holding.

## Hear replies

- To hear one reply, choose **Speak this reply** on the message.
- To hear every reply from a bot, choose its name at the top of the chat, open **Activity**, and
  turn on **Read replies aloud**. The **Voice** next to it overrides your default voice for that
  bot. On mobile, **Read replies aloud** is in the bot's **Chat settings**, under **Advanced**.

## Call a bot

In a chat with one bot, leave the message box empty and choose **Call** (where **Send** usually
sits). If no voice is set up yet, this opens **Settings → Voice** first.

Talk as you would on the phone. You can talk over the bot to cut it off. During a call you can
**Mute**, open the **Transcript**, or minimize the call and keep using the app; **Open call** brings
it back. To end it, choose **Hang up**, or say goodbye and the bot hangs up. Everything said on the
call stays in the chat afterwards.

A call is between you and the bot. To send a bot to a Google Meet, Zoom or Microsoft Teams
meeting, where it takes notes and answers in the meeting chat without speaking, see
[Meetings](/docs/meetings/).

### Live calls

With an **OpenAI** voice connection, calls are live: the bot answers at once in short spoken
sentences, in the bot's **Voice** when the live voices include it.

- When you ask for something that needs real work, such as checking your email, searching the web
  or using the computer, the bot starts it in the background, tells you it's on it, and tells you
  the result when it's done. The full result is posted in the chat.
- If that work needs your approval, the bot asks you on the call; answer yes or no. Passwords and
  keys are never taken by voice: enter them on screen.
- Work started on the call keeps going after you hang up and posts its result in the chat.
- Live calls speak the app **Language** you chose in settings, even if you or the chat switch
  languages.

Without a live voice connection, calls work the regular way: the bot speaks its reply as it writes
it, starting with the first sentence, so you hear what it is checking while it works.

If your server offers live calls without a key of your own, they use your plan's allowance and
credits, like model use. When those run out, calls fall back to the regular way. See
[Account and billing](/docs/account/).

## Bots can call you

In the web and desktop apps, ask "call me at 5 to go over the day", or a bot may call when you've
missed the same reminder several times. You'll see that the bot is calling you; choose **Answer** or
**Decline**. A missed call shows **Call back** in the chat.

## On mobile

Calls work the same way, and live calls play through the speaker. On Android, a call keeps going
with the screen off.

To use your phone's own speech recognition and voices instead, with no key, set **Source** to
**This device** in **Settings → Voice**. Calls then use the phone's voices and are not live. Pick a
**Device voice** for each bot in its **Chat settings**, under **Advanced**.
