Voice
You can talk to your bots two ways:
- Voice call: a hands-free conversation with a bot or a group chat. You speak, the bot's replies are read aloud.
- Dictation: speak into the message box, check the text, then send it yourself.
Both work in the iPhone, Android and desktop apps. The terminal UI has no microphone, so it has neither, but it shows the line a call leaves in the chat (press o on it to read the call).
A call doesn't change the bot. It's the same agent on your computer; you're just talking to it instead of typing.
Start a call
When the message box is empty and the bot isn't working, the send button becomes Call (a waveform). Press it.
The call bar floats at the top of the chat, and the chat stays readable and usable underneath. From left to right:
- the bot's avatar: it pulses while the bot speaks, and tapping it then interrupts the reply
- a line of dots that follows your voice while you speak, ripples while the bot speaks, breathes slowly while it works, and rests as a still row otherwise; while the call connects, a line gliding back and forth takes its place until the voice can really hear you
- the provider's name, whenever audio leaves your device (see "What goes where" below)
- Call settings (the gear), Mute, and End call (the red ✕)
Most bots wait for you to speak. Your lead bot speaks first: once the call connects, it tells you in a sentence or two what deserves your attention, or greets you. If it's busy working, its reply comes when the work is done instead.
What you say on a call, and the bot's replies during it, don't fill up the chat. When the call ends, the chat gets one line instead: a waveform, a short title for what the call was about (such as "Release plan · 03:12"; "Voice chat" until the title is written), and its length. Tap it to read the whole call: the bot, when the call ended and how long it lasted at the top, then what you said, what the voice said, and what went to the bot and back. Your other devices keep the call out of the chat too. If the app closes before the call ends (or loses its connection), the line appears once the call has been quiet for 2 hours.
The bot remembers what matters from the call like from any conversation: its memory picks up what you said, not the voice's wording.
On the iPhone, the first call asks for Microphone and Speech Recognition access. The call keeps going if you leave Codync for the Home Screen, and the bot's Live Activity on the Lock Screen and in the Dynamic Island shows the call, how long it has run and what the bot is doing while you talk, with buttons to mute or end the call without opening Codync.
On Android, the call's notification shows the same: the call's state and what the bot is working on, with Mute and End call.
From Shortcuts (iPhone)
In the Shortcuts app, add Codync's Open Live Mode action and pick a Bot, or leave it empty to call your lead bot, whichever bot you made the lead; if you have none, it asks which bot. With a lead on more than one computer, it calls the first computer's, or the one you pick in Settings → Shortcuts → Lead to call. Running it opens that bot's chat and starts a call. Siri runs it too: say "Talk to Codync" or "Open Live Mode in Codync" to call your lead bot. The computer has to be online; otherwise the action stops with an error instead of starting a call.
From the app icon (Android)
Touch and hold the Codync icon and choose Talk to Codync: it opens your lead bot's chat and starts a call, like Open Live Mode without a bot. With a lead on more than one computer, it calls the first computer's, or the one you pick in Settings → Shortcuts → Lead to call. With no lead, it opens Bots so you can pick one; a computer that can't be reached shows an error instead of starting a call.
Choose a voice engine
Open Voice chat settings from the call bar's gear, or:
- iPhone and Android: tap the connection status at the top of the Bots tab (such as 1 connected), choose Manage computers, and tap Voice chat on the computer.
- Desktop app: Settings → Voice chat.
Under Engine, pick one. The choice is kept on this device. Changed during a call, it takes over that call once you close the settings; the call goes on, still muted if it was.
| Engine | Needs | How it works |
|---|---|---|
| On device (default) | Nothing | Your device turns speech into text and reads replies with the system voice |
| OpenAI | Your OpenAI API key, kept on the computer | OpenAI's models hear you and speak |
| Gemini | Your Google AI Studio key, kept on the computer | Google's models hear you and speak |
On device
- You speak in the language set under Language (by default, your device's first language).
- A pause ends what you said and sends it to the bot as part of the call. If the bot is busy, it's queued like any other message. Set the pause under Send after a pause of (1 s, 1.5 s or 2.5 s).
- Each reply is read aloud in the language it's written in, skipping code blocks and formatting. Set Reading speed to Slower, Normal or Faster. The microphone pauses while a reply is read.
- When the bot asks for approval or asks you to choose, the call tells you what it asks; answer it in the chat. When the question comes from a thread, the call names the thread. If it runs into a problem, the call tells you that too.
- A reply that arrives while the call briefly loses its connection is read once it's back.
- If speech recognition keeps failing, the call stops and asks you to check the microphone and speech recognition settings.
In the desktop app, on-device speech is available on macOS only. On Windows and Linux, calls use OpenAI or Gemini; without a key, the call asks you to add one in call settings.
OpenAI or Gemini
Each provider has two modes, under Mode:
- Realtime: a live conversation with the provider's realtime model. You can talk over replies.
- Speech: the same loop as On device, using the provider's speech-to-text and text-to-speech models instead. What you say is sent when you pause, and replies are read aloud.
In Realtime mode, the voice model is the bot's voice line, not the bot: it talks with you, and the bot does the work. It passes your requests on to the bot (along with your own words, so nothing gets lost in its wording), reads the bot's replies out in shortened form, keeps you posted while the bot works (and tells you what it's doing when you ask), lets you know when an approval is waiting, answers an approval for you once you've said which option, can stop the bot (which also stops its background work, waiting threads included), and can look up what the bot remembers.
Before a Realtime call, your computer gives the voice model the background it needs: the date and time, what your bots remember about you, this bot's settings and memory, the project's git branch and recent changes, what the bot is working on, the recent messages and calls in this chat, your other bots, and the files shared in the chat. When the call reconnects or you switch engine mid-call, the voice model also gets what was said on the call so far. Like your audio, it goes to OpenAI or Google on your own key.
Below the key you can pick the Model (or, in Speech mode, Speech to text and Text to speech) and the Voice. The default models follow each provider's newest ones. This month shows an estimate of the minutes used, counted on each device. The provider bills you; it isn't a bill.
If the connection drops mid-call, the bar shows the connecting line with "Reconnecting" and the call picks up where it left off, usually within a few seconds, without hanging up. A phone call, Siri or another app taking the audio pauses the call; it picks up again when that ends, or tap Audio paused by another app. Tap to resume. under the bar. On the iPhone and the computer, plugging in or removing a headset moves the call to the new microphone; on Android, removing the headset pauses the call until you tap to resume.
If a provider can't start, or the connection doesn't come back after a few tries, the call switches to on-device voice without hanging up and says so under the bar. (In the desktop app this happens on macOS, where on-device speech exists.)
Add your API key
Keys are entered on the computer, never on the phone:
- In the desktop app on that computer, open Settings → Voice chat. For a computer you reach over SSH from the desktop app, use the Voice chat button (waveform) on its card in Settings → Computers & devices.
- Pick OpenAI or Gemini as the engine. Use Get a key from OpenAI (or Gemini) if you need one.
- Paste the key and press Save. The computer checks it with the provider and keeps it only if it works; otherwise you see the provider's error.
Once saved, the section shows Key saved with Test key, Replace key and Remove key. The iPhone shows whether the computer has a key and can test it, and tells you where to add one if it doesn't.
Use your own speech server
The OpenAI section has a Server row. It reads api.openai.com until you type the address of an OpenAI-compatible server of your own there, such as a local speech-to-text and text-to-speech server (http://127.0.0.1:8890). The address must be https, or http on the computer itself. The computer checks the saved key with that server before it keeps the address, and from then on your recordings and spoken replies go to that server instead of OpenAI. Such a server works in Speech mode only, so Mode then reads Speech on every device, and the mode you had picked comes back when you clear the row to go back to OpenAI. Speech to text and Text to speech then list every model the server offers, whatever it calls them; changing the server, or clearing it, sets them back to the new one's defaults. If the saved key doesn't work with the server, remove the key, save the server, then add the server's key. Clearing the row also removes the saved key, so a key meant for your server never goes to OpenAI: add your OpenAI key again to use OpenAI.
Like the key, the server is set in the desktop app on the computer itself. The iPhone and Android apps show which server the computer uses. A Codync Cloud computer always uses OpenAI.
The key is stored encrypted in Codync's vault on the computer (with the system keychain, as for connector credentials), so one key serves the iPhone and the desktop app. It's never logged and never sent to the relay or to Codync's cloud.
What goes where
| You use | Your audio | Your key |
|---|---|---|
| On device | Stays on your device. Only the recognized text goes to the computer, as a message. Apple's speech recognition may use Apple's servers when on-device recognition isn't available for your language. On Android, calls are always recognized on the phone; dictation uses the phone's speech service when the phone can't recognize your language itself, and that service may use its provider's servers (Google's on most phones). | Not needed |
| Realtime | Streams directly between your device and the provider (OpenAI over WebRTC, Gemini over a WebSocket). It doesn't pass through your computer or the relay. The written transcript of the call is kept on your computer. | Stays on the computer, which gives your device a short-lived credential for each call |
| Speech mode, or dictation with a provider | Recordings go to your computer over the encrypted connection, and the computer sends them to the provider, or to the server of your own you set for the OpenAI engine, with the names of the bots on the computer as spelling hints. The spoken replies come back the same way. | Stays on the computer, which sends it with each request |
The relay only ever sees encrypted data. Whenever audio leaves your device, the call bar names the provider (the engine, OpenAI, also when it uses your own server).
Dictation
The microphone button (Dictate) sits beside the send button. Press it and speak: the words go into the message box, after anything you've already typed. Nothing is sent until you send it.
While it listens, the microphone becomes a recording pill with a stop button, the time, and dots that follow your voice.
Pick the dictation engine
The small arrow beside the microphone (Dictation engine) chooses On device, OpenAI or Gemini. It's kept on this device, separate from the call engine.
- On device: words appear as they're recognized, in the language set for calls. On the iPhone and Android, stop keeps what was recognized so far, and dictation tells you why if it can't start or stops on an error; a pause with nothing said just ends it. In the desktop app, dictation keeps listening through pauses until you press stop.
- OpenAI or Gemini: Codync first checks that the computer has the provider's key (if not, voice settings open so you can configure it before anything is recorded). The pill appears once the microphone is open, so nothing you say is lost. Nothing is typed while you record. When you press stop, the pill shows "Transcribing…" and the text lands in the box when it arrives. It uses the Speech to text model from Voice chat settings, and the recording counts toward that provider's minutes this month.
If dictation uses a provider that your calls don't, Voice chat settings show a separate section for it ("For dictation in the message box.") with its key status and speech-to-text model.
Sending while dictating sends what you typed and what you said together: with OpenAI or Gemini, and in the desktop app, it waits for the last words first; with On device on the iPhone and Android, it sends what was recognized so far, as stop does. If the transcription fails, nothing is sent and what you typed stays in the box. Starting a call or switching chats drops a dictation that is still recording or transcribing. A call takes the microphone.
In the desktop app on Windows and Linux, the microphone works with OpenAI or Gemini; On device is hidden, and an earlier On device selection uses OpenAI instead.
Voice input shortcut (Mac)
In the desktop app on macOS, Settings → General → System → Voice input shortcut lets you dictate from any app: Right ⌘ twice, Left ⌘ twice, or Off (the default).
Double-tap the key and dictation starts in the chat window's message box (your lead bot's chat when none is open, or the top bot's if you have no lead). Tap the key once more to stop and send what you said, once the last words are in. Press the pill's stop instead to keep the words without sending. If the notch island is on (Settings → General → System → Show in Notch), dictation goes into the island's message field instead. During a call the shortcut does nothing.
The shortcut needs no Accessibility or Input Monitoring permission, and the app never sees what you type.
Google Search during Gemini calls
Choose Gemini and Realtime in Voice chat, then turn on Google Search to let the voice search public web information and answer you directly. Requests about your code, files, computer or project still go to your bot. The switch is off by default and saved on this device. Your key needs Google Search quota, separate from Live API access. If Google refuses a search-enabled call for quota, turn off the switch to try voice without search. Google may charge for searches in addition to the voice call; see Google's pricing.
During a call, the change applies when you close settings. It does not affect Speech mode or OpenAI calls. The switch appears only when the computer supports it; other computers keep Gemini calls working without search. When Google supplies search suggestions, they appear with the answer below the call bar. Their links open in your browser. Suggestions are kept only during the call; they are not added to its saved transcript.