How-to guides
How to talk with GPTree by voice
Start a voice session and just talk. GPTree replies straight away, keeps listening while your AI model writes the answer, then gives you a short spoken summary. With Live voice it answers as you talk and you can cut in at any time. Say "read it" for the whole answer. You can also dictate a message or hold the mic for a single command. Step by step for the browser, the Mac app and the phone apps.
There are three ways to use your voice in GPTree:
- A voice session is a real back-and-forth. You talk, GPTree replies out loud, and it keeps listening until you end it.
- Dictation writes what you say into the message box and waits for you.
- A single command does one thing in the app, such as opening a conversation, and then stops listening.
Start a voice session
Any of these starts one:
- Click the mic in the top bar.
- Click the waveform button next to the mic in the message box.
- Tap Option Space on a Mac, or Ctrl Space on Windows and Linux. A quick tap starts a session; tap again to end it.
A dark voice bar takes the place of the message box, with a glowing orb that moves as you speak. Just talk. When you pause, GPTree works out what you meant:
- A question, or something to write or talk through, goes to your own AI model in the conversation on screen. GPTree tells you straight away that it has asked, for example "Sure, I'll check that in Q4 pricing.", and listens again while the answer writes itself in the conversation. When the answer is ready, GPTree gives you a short spoken summary at the next pause in your conversation. Say "read it" to hear the whole answer.
- Something to do in the app, such as "open billing" or "branch that answer two ways", is done straight away, and the bar says what happened.
- Something quick, such as a greeting or a question about how GPTree works, is answered by the voice itself, without waiting for your AI model.
You never have to wait for an answer to finish before you speak again: ask a follow-up, ask something else, or give a command while it writes. If no conversation is open, your first question starts a new conversation and is answered there, so the whole exchange is saved like any other conversation you can branch later.
Voice keeps track of what you said a moment ago, so you can correct yourself ("sorry, I meant the Graft agent") or follow up ("open that first") without saying it all again. When a follow-up goes to your AI model, the earlier words it builds on go with it. If you ask a question while you are on another page, such as your settings, voice opens your conversation and asks there.
Live voice and Standard voice
A voice session runs on one of two voice engines:
- Live voice hears you and answers as you talk, usually within a second, in one natural voice for the whole session. You can cut in at any time just by talking. It hands questions to your own AI model in the conversation and anything you ask it to do to GPTree, then tells you in its own words when the answer is ready or the action is done. Live voice comes with voice on every plan that includes voice, and uses credits for every second the session is on, plus the credits your answers use as usual.
- Standard voice turns your speech into text with your browser or device first, then replies. It is used when Live voice is not available, when your browser or device cannot run it, when Speak replies is off, or when Live voice cannot connect.
There is no setting for this: every voice session uses Live voice when it can. If Live voice cannot start or drops and cannot reconnect, the session carries on with Standard voice and the voice bar says so. What Live voice sends to OpenAI is explained in Voice, your microphone and your privacy.
Answers you asked for
Each question you ask by voice gets a small chip above the voice bar:
- Waiting: the conversation is still writing an earlier answer, so your question is next in line. Stop takes it back.
- Writing: your AI model is writing the answer. Stop stops it.
- Ready: the answer is in the conversation. Read reads the whole answer aloud, the same as saying "read it".
- Failed: the answer did not finish, and GPTree tells you why. Retry asks the same question again.
- Stopped: you stopped it.
A summary waits for a quiet moment: it never talks over you, and anything you say first always comes first. In the transcript, lines from the voice are labelled Voice, and summaries carry the name of the model that wrote the answer.
When an answer asks you something, such as a quiz question or "which one do you want?", voice reads the question to you as it was written and never answers it for you. Just say your answer: it goes back to the same conversation, so you can work through a quiz out loud one question at a time.
While you talk
- Interrupt by tapping the orb, or by pressing Space on a keyboard. The voice stops and GPTree listens; an answer being written keeps going in the conversation. With Live voice you can also just start talking.
- Take your time: a short command goes as soon as you stop, but once you are saying something longer GPTree waits longer before it answers, and longer still if you stop mid-sentence, so you can pause to think without being cut off.
- Send now: if you have finished a sentence and do not want to wait for the pause, tap the orb.
- Read it: say "read it" to hear the latest answer in full, "read the summary" to hear its summary again, or "say that again" to repeat the last thing GPTree said.
- Cancel that question: say "cancel that question" or "stop the answer" to stop the answer being written. Saying "never mind" right after you ask does the same.
- Never mind: say "never mind" to drop what you just said and keep going.
- Mute closes the mic without ending the session. Tap Unmute, or the orb, to carry on.
- Captions only: tap Voice on to switch to captions only. Replies and summaries are then written out but not spoken.
- Transcript shows what you said and what GPTree did this session.
If you go quiet for about a minute, the session pauses itself and the mic closes. Tap the orb to pick up where you left off. A paused session ends by itself after a few minutes. Switching to another tab or app also pauses it, and nothing is read out while you are away.
End a voice session
Click End on the voice bar, press Esc, or just say so in your own words: "end session", "stop voice", "turn voice off" and "that's all" all work. Saying "stop" on its own also ends it. A question that only mentions the session, such as "what does a voice session cost?", is answered instead.
Ending the session does not stop answers that are still being written: they finish in their conversations as usual.
Dictate a message
- Tap the round mic button next to Send. A green bar appears in the message box with a live level meter.
- Talk. Your words land in the draft as you speak, after anything you had already typed.
- When you are done, say "send it", tap Send, or tap Done to stop dictating and keep editing.
Dictation never sends by itself. Say "scratch that" to throw away what you just dictated and keep what you had typed before. After a few seconds of silence, dictation stops on its own. Starting dictation ends a voice session, because both need the mic.
Give a single command
- Hold the mic in the message box, say the command, and let go.
- Or hold Option Space on a Mac (Ctrl Space elsewhere) and release when you finish.
The voice bar shows Listening, then Working, then the result, written out and read aloud if Speak replies is on. A question asked this way gets the same quick reply, and its chip shows until the answer is ready. The summary is then spoken, as long as voice is not busy with something else and Do not disturb is off.
Examples to start with:
- "Open my pricing conversation."
- "Start a new conversation." Add what it is for, such as "start a new conversation about our launch plan", and that becomes its first question.
- "Show me all my conversations." Opens the list of your conversations.
- "Go to the team tab." On an Agents page, a tab's name opens that section: Team, Needs you, Work, Artifacts, Decisions or Meetings. Anywhere else, say "the agents team" or "what needs me".
- "Scroll to the top." or "Scroll to the bottom." Moves through the conversation on screen.
- "Branch that answer three ways: optimistic, skeptical, one for the board."
- "Open the skeptical branch." A branch of the conversation on screen opens as a pane beside it.
- "Switch this pane to Claude Opus."
- "Read me the last answer."
If you ask for a branch without saying what it should explore, such as "branch from my first message", voice opens it empty and asks what it should explore. Your answer becomes the branch's first message.
Say "What can I say?" to see every command available to you on the current screen.
Answer a question voice asks you
- Which one? If a reference is unclear, voice offers two or three options. Say one by name or by position, such as "the second one", or tap it. If you answer with something else, voice carries on with what you asked instead of asking the same question again.
- Confirm. Before anything lasting, such as sharing a conversation publicly, voice asks and shows the word to say, for example "share". Say it or tap the button. Silence for ten seconds cancels.
- Confirm delete. Deleting needs the word "delete" itself, or a tap on the confirmation. "Yes" is not enough.
The voice you hear
Replies are read by a natural voice. To make the audio, the reply text is sent to OpenAI's voice service, and the audio uses credits like any other reply. The words appear on screen as they are spoken. If you would rather use your device's built-in voice, turn off Natural voice in the Voice panel. When the natural voice cannot be reached, that reply is read by your device's voice instead; if it happens again soon after, or you are out of credits, the device voice stays for a while so the voice does not keep switching back and forth.
Writing a spoken summary of a long answer also uses a small amount of credits. A short, plain answer is read as it is, with no summary needed.
With Live voice, the live voice speaks in the same natural voice for the whole session and says results in its own words, keeping names and numbers as they are. Ask it to read an answer and it reads it word for word.
On the Mac app
The Mac app works like the browser, plus one option: in the Voice panel, turn on Push to talk in any app to use Option Space while GPTree is in the background. It is off by default, because while it is on no other app can use Option Space (some launchers use it). If another app already holds the shortcut, GPTree tells you and keeps it inside its own window.
On iPhone and Android
Tap the waveform button at the end of the message box (it shows while the box is empty) to start a voice session. The full-screen voice view shows the orb, what GPTree is saying, and your last exchange. Tap the orb to interrupt or to carry on after a pause. At the bottom:
- Aa lets you type instead of talking.
- The large square button ends the session.
- The mic button mutes and unmutes.
CC at the top switches to captions only. Swipe the handle at the top down to shrink the session to a bar over the message box; tap the bar's expand button to bring it back.
Tap the mic in the message box to dictate, exactly as above.
Change how voice behaves
Open the Voice panel with the small arrow next to the mic in the top bar:
- Start a voice session: the dark card at the top starts or ends one.
- Speak replies: off means captions only.
- Natural voice: off uses your device's built-in voice instead.
- Do not disturb: no spoken updates, for example when a branch voice made finishes writing or an answer you asked for is ready. The answer's chip shows instead, and "read it" still works.
- Keep what I said: off keeps a record of what voice did but not your words. See Voice, your microphone and your privacy.
- Voice history: your recent voice sessions, newest first. Open one to read what was said and done, or delete it. A session also shows as a Voice session card in the conversation it happened in; only you see your own sessions.
If the mic does not appear or does not hear you
- No mic at all: your browser may not offer speech recognition to websites (Firefox does not). Try Chrome, Edge or Safari, or the GPTree app.
- "Microphone access is off": allow the microphone for GPTree in your browser's site settings, or in your phone's Settings, then tap the orb or Try again.
- The session paused: it pauses after about a minute of quiet, or when you switch away. Tap the orb to continue.
- Live voice did not start: the voice bar says why, for example "Standard voice: Live voice is not available right now.", and the session carries on with Standard voice. The next session tries Live voice again.
- A single command stops straight away: it ends when it hears nothing for a while. Hold the mic or the keys while you think, and let go when you have finished.