Privacy policy

• No account, no sign-up. We don’t know who you are, and we don’t keep a profile of you.
• Your history stays on your device. So do your Dictionary, snippets, notes and settings. We never get a copy we can read.
• Your last recording stays only on your device, for up to 24 hours, so you can recover that dictation if something goes wrong. Each new dictation replaces it, and “Delete it now” deletes it anytime.
• No AI provider keeps your voice or your words. To turn your speech into polished text, your recording and your words are sent to our AI providers with zero data retention: they’re used once to answer you, then not kept — not by us, not by them — and never used to train AI. (The live words you may see while you talk come from your device’s own speech recognition, by Apple or Google — see below.) Murmur asks before it sends anything to these AI services, and you can change your answer anytime in Settings.
• No ads, no analytics, no tracking. We never sell or share your data.
• Your pictures stay on your device. The only image that ever leaves it is a screenshot you choose for “Reply from a screenshot” — sent only to write your replies, and never kept.
• Files you transcribe stay on your device. Their sound is sent in short pieces only to be written down, with zero data retention, and nothing about the file is kept — only the transcript, in your History on your device.
• A meeting you transcribe is recorded only after you press Start, with a red dot and a timer on screen. The sound goes in short pieces to be written down, with zero data retention, and is never stored — not by us, and not on your device. The transcript and notes stay in your History on your device.
• The only things we keep are things you choose to send us: a suggestion or support question, and — only if you turn them on — voice samples, which you can delete anytime.
• Moving your words to another device? They’re encrypted on your device first, so we can’t read them, and they’re deleted as soon as they’re opened.

Murmur is currently operated by Kristen Closson, an individual, until the company that will run Murmur is formed. In this policy, “Murmur”, “we” and “us” mean the operator. This policy covers the Murmur website and web app, the Murmur iPhone app and keyboard, and the Murmur apps for Mac and Windows.

Questions? Use the Contact us form on the Support page or the Suggest box in the app.

Everything you save in Murmur lives only in Murmur’s storage on your device: your name (if you share it during setup), history, Dictionary (including the words Murmur learns from your fixes), snippets, quick sends, Scratchpad notes, stats, training progress, settings, and what setup learned about how you talk (a few numbers — see Talking with Pip). We never receive a readable copy. There are no accounts, so nothing syncs between your devices on its own — each one keeps its own, unless you move your words with a one-time code.

Web app

Your browser’s storage for this site (localStorage, and IndexedDB for your last recording — kept up to 24 hours), plus a cache of the app’s pages so it opens offline.

iPhone

The app’s storage, and a shared space between the app and the Murmur keyboard (an “App Group”) for what the keyboard needs: a copy of your Dictionary, its settings, your most recent dictation, the last keyboard dictation’s recording (kept up to 24 hours), words you fixed after a dictation, and words you chose to keep exactly as you typed them.

Mac and Windows

The app’s storage (including your last recording, kept up to 24 hours), a settings file (your shortcuts and preferences), and a small troubleshooting log on your computer (times, character counts, permission states and the names of the apps Murmur typed into — never your words). The log stays under about half a megabyte, and Murmur clears out anything in it older than 14 days whenever it runs. Nothing reads that log but you.

Delete the iPhone app, or clear this site’s data in your browser, and it’s all gone. Anywhere, Settings → Clear all local data wipes your name, history, Dictionary, snippets, quick sends, Scratchpad, stats and your last recording in one tap. (On Mac and Windows, removing the app can leave its settings file and troubleshooting log behind; you can delete them yourself.)

Read it back reads your text aloud with one of your device’s own voices (the one you pick in Settings, or the best one there), right on your device. Nothing is sent anywhere to do it.

Transcribe a file keeps only the transcript, in your History on your device, named after the file. The file and its sound aren’t kept anywhere. On iPhone, a file you share to Murmur is copied into the app’s private storage on your phone only while it’s being transcribed, and deleted as soon as you’re done — or the next time you open Murmur.

Ask about this and Draft a follow-up email keep your questions, the answers and the email draft only in History on your device, with the transcript they’re about. Delete the transcript from History and they go with it.

A meeting you transcribe keeps its transcript and notes in your History on your device, saved as the meeting goes on. The meeting’s sound isn’t kept anywhere — see Transcribe and note a meeting.

The words you see while you talk

While you talk, your device’s own speech recognition may show your words live. That’s handled by whoever makes it, under their own privacy policy: Apple’s speech recognition in the iPhone and Mac apps and in Safari (on your device when it supports that, otherwise on Apple’s servers), and Google’s in Chrome. Firefox and the Windows app don’t show live words.

The final transcript

When you finish, your recording goes to our server, which passes it to our speech-to-text provider — xAI, through Vercel AI Gateway — with zero data retention, to turn it into text. That’s how Murmur catches every word, even when you pause. Your Dictionary words go along as spelling hints. Our server passes the recording along, gets the text back, and doesn’t keep either one. Neither Murmur nor our AI providers keep your audio or its text, and neither is used to train AI models. We only use a speech-to-text service that accepts zero data retention; if it ever refuses, your recording isn’t sent at all.

On iPhone, “Keep my voice on this iPhone” changes this for all dictation, in the keyboard and in the app — see iPhone. The Mac and Windows apps and the web app have no such switch: there, your recording always goes to speech-to-text as described above, with zero data retention.

Recovering your last dictation

Your last recording stays only on your device — never on our servers — for up to 24 hours, so you can recover that dictation if something goes wrong: speech-to-text or the polish fails, your connection drops, or the app closes before your words come back. Tap “Recover last dictation” (in Recent, or in the Murmur keyboard’s ☰ menu on iPhone) and it goes through speech-to-text and cleanup again, as described here. Each new dictation replaces it, and it’s deleted automatically after 24 hours. Want it gone sooner? Tap “Delete it now”.

Nothing is kept when there’s nothing to recover — a tap that recorded nothing, or silence that produced no words — and the Murmur keyboard never keeps a recording from a password box. With “Keep my voice on this iPhone” on, nothing is recorded, so there’s nothing to recover.

Transcribing a file

When you transcribe an audio or video file, the file itself never leaves your device. Murmur decodes its sound on your device — in the iPhone app, the app does this on your phone — and sends it in short pieces of about a minute to the same speech-to-text service as your dictation, with zero data retention. Silent stretches aren’t sent at all.

Neither Murmur nor our AI providers keep the file, its sound or its text. A file is never kept for “Recover last dictation” either — that only ever keeps a dictation you recorded. If you Polish, Translate or Summarize the transcript, ask a question about it, or draft a follow-up email from it, its text (never the sound) is sent to our AI provider like any other text, as described below.

Words it isn’t sure about

To spot words it may have misheard, Murmur compares the live words with the final transcript, right on your device, and gives the words they disagree on a dotted underline. When you fix one (or fix a word any other way), the new spelling is saved to your Dictionary — on your device, like everything else you save.

When you transcribe a meeting, Murmur writes it down while it happens — who said what, and when — then makes notes from it. Here’s exactly what that involves.

What’s recorded

• Nothing until you press Start, and nothing after you press Stop (or after two hours, when it stops by itself). While it records, the screen shows a red dot and a timer. In the Mac and Windows apps, the Murmur icon in the menu bar or system tray shows a dot too, with a Stop item at the top of its menu.
• A call, in the Mac and Windows apps: your computer’s sound — the call, so what the other people say — and, if you choose, your microphone, for your side. On a Mac this needs the “System Audio Recording Only” permission, which macOS asks for the first time you start.
• A meeting in the room, on any device: your microphone.
• A meeting in another browser tab (Chrome or Edge on a computer): only the sound of the tab you choose to share — never its picture — and your microphone, if you choose.
• Transcribe what’s playing (on a computer) uses this same recorder for a video, a talk or a podcast, and its sound is sent and never kept the same way. In the Mac and Windows apps it hears your computer’s sound with your microphone off; in Chrome or Edge, only the tab you share (or your microphone, if you choose); anywhere else, your microphone, hearing it play out loud. Only its transcript is kept, and a summary is made only when you ask for one.
• “You” and “Others” come from where the sound came from: your microphone is You, your computer’s sound is Others. Murmur doesn’t recognize voices and makes no voiceprints.
• To offer to transcribe a meeting, the Mac and Windows apps notice when another app starts using your microphone — only that it’s busy, and which app (like Zoom or Google Chrome) — and show a banner at the top of your screen. That check happens on your computer every few seconds (never while a meeting is being recorded), nothing about it is sent, and it never starts recording by itself. Turn it off in Settings with “Offer to transcribe a meeting when one starts”.
• On iPhone, with “Keep my voice on this iPhone” on, a meeting isn’t recorded at all, because no sound may leave your phone.

What’s sent

• The sound, in short pieces (a minute at most) while the meeting goes on, to the same speech-to-text service as your dictation — xAI, through Vercel AI Gateway — with zero data retention, only to write it down. Silent stretches aren’t sent.
• The transcript’s text — never the sound — to our AI provider, Anthropic’s Claude, through Vercel AI Gateway, with zero data retention, to make the notes; and, only when you use them, to answer your questions about it or draft a follow-up email.
• Neither Murmur nor our AI providers keep the sound or its text, and neither is used to train AI models.

What’s kept

• The transcript and the notes, in your History on your device — saved as the meeting goes on, so what’s already written down survives a reload or a crash. Rename them or delete them anytime.
• The sound is never stored — not on our servers and not on your device. Each piece lives only in Murmur’s memory until its words are back, then it’s gone. A piece that can’t be sent waits in memory to be tried again, and goes when you close the notes. So if Murmur closes in the middle of a meeting, the last minute or so that hadn’t been written down yet is lost, and the notes say so.
• A meeting’s sound is never kept for “Recover last dictation” and is never used for voice samples.

Telling people

You’re responsible for letting everyone know you’re transcribing the meeting, and for following the recording laws where you and they are — some places require everyone’s permission before a conversation is recorded. Before each meeting, Murmur reminds you and gives you a line to copy and send (turn that off with “Remind me to tell people” in Settings).

To polish your words, the text — never the audio — is sent to our AI provider, Anthropic’s Claude, through Vercel AI Gateway, along with what it needs to do the job (your style, cleanup level, Dictionary words, and the language to “Write in”). It’s used only to send back the result, and Murmur doesn’t store it. Every request is sent with zero data retention: our AI provider doesn’t store it and doesn’t use it to train its models.

The same goes for everything else Murmur’s AI does with your text:

• “Say what you mean” (writing the message you described)
• voice edits (“Pip, make that shorter”, “change Tuesday to Wednesday”) and the tone buttons — Warmer, Shorter, Firmer, Funnier — which send the text being changed and your instruction
• “Write in” translation while you dictate
• “Summarize” in the Scratchpad
• Polish, Translate and Summarize on a file’s transcript, which send its text in parts
• the notes for a meeting (Transcribe and note a meeting), which send its transcript’s text — labeled You and Others, never the sound — in parts when it’s long
• “Ask about this” and “Draft a follow-up email” on a transcript (a file’s, a meeting’s, or a long dictation’s), which send the transcript’s text — never its sound — with zero data retention, the same as Summarize. Ask also sends your question, your last few questions and answers about it, and your Dictionary words. The email also sends your Dictionary words and the name you gave during setup (to sign it), and uses your Summarize notes instead of the transcript when you made them. The answers and drafts are kept only in History on your device.
• a little of the text just before your cursor (up to about 400 characters, which can include words you typed yourself), sent with a dictation so the new words fit what came before — read-only, used only as context, never changed or kept
• “Polish all” in the iPhone keyboard, which sends the whole text box you ask it to polish
• text you share into Murmur from another app (the Share menu on Android), polished like a message you typed — until then it stays on your device, and it’s never put in a web address, so it can’t end up in request logs
• your answers during setup, used only to pick out the names and words to save in your Dictionary — on your device — and to write Pip’s short reply to them

In the Murmur app, set cleanup to “Off” and your dictation isn’t sent for polishing at all. Keyboard dictation on iPhone is always polished, so its text is always sent for cleanup.

Setup is a short conversation: Pip, Murmur’s companion, asks a few questions — your name, who you talk to, what keeps you busy, something you’re looking forward to, any pets — and you answer out loud (or type). It never asks about health, religion, politics, money or relationships, and skipping is always fine.

Your answers are handled like any dictation: a spoken answer goes through speech-to-text with zero data retention, and its text goes to Anthropic’s Claude with zero data retention — the same as cleanup — to pick out names and words for your Dictionary and to write Pip’s short reply, which is shown in Pip’s speech bubble. Nothing is saved until you check it, and then only on your device. Murmur doesn’t store your recordings or your answers.

Pip’s voice is pre-recorded. It’s an AI-generated voice that we recorded once, ahead of time: every line Pip says out loud — the greeting, the questions, and short reactions like “Oh, nice!” — is the same for everyone. Nothing you say or type is sent anywhere to make Pip’s voice. Tap the speaker button to mute Pip (remembered on this device); everything Pip says is shown as text either way.

What Pip learns about how you talk — how fast you talk, how long your pauses run, whether you speak softly, and which language you speak — is worked out on your device, from the microphone’s level and the words that came back. Pip shows you what it learned, each as a note or a setting you can switch on (like how long the iPhone keyboard waits through a pause). Only those numbers and the settings you keep are saved, on your device.

Setup keeps no recording. The one exception is yours to make: if you’ve already turned on voice samples, setup asks once whether to send your answer about what you’re looking forward to as a sample, and sends nothing unless you tap Send.

Dictating with the Murmur keyboard

By default (for the best accuracy), when you finish a dictation — in the keyboard or in the Murmur app — Murmur sends the recording for speech-to-text as described above. A keyboard dictation’s recording is saved on your phone while it’s sent; the most recent one then stays on your phone for up to 24 hours, so you can recover it from the keyboard’s ☰ menu or the Murmur app (see Recovering your last dictation), and is then deleted.

Prefer your voice never leaves your phone? Turn on “Keep my voice on this iPhone” (in Settings or the keyboard’s ☰ menu). Then every dictation on your iPhone — in the keyboard, and in the Murmur app itself (its mic button, “Ask Murmur”, the Action Button, Back Tap and “Text with Murmur”) — is turned into words by Apple’s speech recognition alone, and Murmur doesn’t save or send any recording. In the Murmur app that always happens entirely on your phone: if your iPhone can’t do it there for your language, Murmur tells you and doesn’t dictate, rather than send your voice anywhere. The keyboard does it entirely on your phone on supported iPhones. Your text is still sent for AI cleanup, and voice samples are never offered.

Stay ready, and the Live Activity

After each keyboard dictation, Murmur keeps the microphone ready for a few minutes (5 by default; you can choose 15 or 60, or end it anytime from the keyboard or the app) so your next dictation starts instantly. Your iPhone shows its orange microphone dot the whole time. Rather it didn’t? Turn off “Every time” in the keyboard’s ☰ menu (or “Keep the mic ready after I talk” in Settings), and the microphone turns off the moment each dictation ends. While it waits, Murmur ignores the sound completely — nothing is transcribed, recorded, stored or sent until you tap to talk. A phone call or Siri ends it, and it switches off on its own when the time is up.

While the session is open, Murmur shows a Live Activity on your Lock Screen and in the Dynamic Island with the last few words you dictated and a word count (and, for a few seconds after a voice command, what it did — “Added to Reminders · 3:00 PM”) — so anyone who can see your screen can see them. It’s made on your phone and isn’t sent anywhere.

Reminders, calendar events and notes by voice

Start with “Pip” (or “Murmur”) and say what to do — “Pip, remind me to call Sam at 3”, “Pip, add lunch with Dana Friday at noon to my calendar”, “Pip, note: pick up the dry cleaning” — and Murmur does it instead of typing it. What you asked for is worked out on your phone: the command isn’t sent anywhere to be understood (your recording goes for speech-to-text just like any dictation, as described above). Reminders and events go only into your iPhone’s own Reminders and Calendar apps — never to Murmur’s server — and nothing is sent to anyone: an event Murmur adds has no invitees. Notes go to your Scratchpad. Each one can be undone right away.

Murmur asks for permission the first time you use one. Calendar access is add-only: Murmur can add events but can’t see your calendar. For Reminders, iPhone only offers full access; Murmur only adds the reminders you ask for (and removes one if you tap Undo) and never reads or sends the others. If you say no, Murmur saves what you asked for to your notes instead. You can change either anytime in iPhone Settings → Murmur.

What the keyboard does with what you type

The Murmur keyboard reads the text around your cursor to autocorrect, capitalize, place your dictation and notice when you fix a word it just wrote — all on your phone. It doesn’t log your keystrokes and doesn’t read your clipboard (it only copies a dictation there when it can’t type it in). Nothing you type is sent anywhere, with one exception you start yourself: a voice edit or tone button in the keyboard sends the text it’s changing — which can include words you typed — for the AI edit, with zero data retention.

Next-word suggestions are worked out on your phone, from a word list built into the keyboard and the phrases you type (kept on your phone only; Dictionary → Reset keyboard predictions clears them).

When you fix a word right after a dictation, the keyboard remembers the new spelling on your phone and adds it to your Dictionary next time you open the app — so, like every Dictionary word, it’s sent along with your dictations as a spelling hint. Words you keep exactly as you typed them (by undoing an autocorrect) stay on your phone.

Siri, Shortcuts and notifications

• “Text with Murmur”: the person Siri or Shortcuts passes along stays on your phone; it’s only used to open Messages with your polished draft. You press Send.
• “Polish my clipboard” and “Translate my clipboard”: read your clipboard only when you run them, and send that text for cleanup or translation like any other text.
• “Murmur Note”: your note is polished like any dictation and waits on your phone until the app adds it to your Scratchpad.
• Reminders: if you allow notifications, Murmur may send a friendly daily nudge to dictate. It’s scheduled on your phone, never contains your words, and you can turn notifications off anytime in iPhone Settings.

The desktop app runs the same Murmur web app inside a window, so everything above applies. On top of that, to work in any app:

• Typing your words in. Murmur pastes your text where your cursor is, through the clipboard. It marks that paste so Windows clipboard history, Windows’ cloud clipboard and Apple’s Universal Clipboard skip it — and so do clipboard managers that respect the standard markers — then, a moment later, puts back what you had copied before (unless you’ve copied something new since). You can turn that off in Settings; your dictation then stays on the clipboard, still marked. “Copy last” is a normal copy, since you asked for it.
• Which app you’re in. Murmur reads the name of the app you’re typing in to pick a writing style. That happens on your computer; only the style is sent.
• Editing in other apps. When you start a voice edit or tap a tone button for text in another app, Murmur reads the text you highlighted (or the words it just typed) and sends it for the AI edit, with zero data retention. Before replacing its own words, it checks the text just before your cursor — on your computer only.
• Shortcuts. To notice hold-to-talk and your other shortcuts, Murmur sees which keys are pressed. It acts only on its shortcuts and never records, stores or sends what you type.
• Transcribe and note a meeting. Only after you press Start, the app records your computer’s sound (the call) and, if you choose, your microphone, and hands them to Murmur’s page a second at a time, in memory — nothing is written to disk. While it records, the menu bar or tray icon shows a dot, with a Stop item at the top of its menu; it stops by itself after two hours. On a Mac, Murmur’s own sounds are left out of the recording. To offer to transcribe, it notices when another app starts using your microphone — only that, and which app. See Transcribe and note a meeting.
• No extras. The desktop app has no analytics, no crash reporting and no automatic updater — it only talks to Murmur’s own server.

When you use “Reply from a screenshot”, the screenshot you choose — made smaller on your device first — and what you type or say about what to reply are sent to our AI provider (Anthropic’s Claude, through Vercel AI Gateway) only to write your reply suggestions, with zero data retention. Murmur never stores the screenshot or the replies: they stay in the app’s memory while that screen is open and are gone when you close it. If you say what to reply instead of typing it, your recording goes through speech-to-text like any dictation.

On iPhone, sharing a screenshot to Murmur saves a smaller copy in the app’s private storage on your phone until the Murmur app opens it; it’s deleted as soon as the app takes it (or, if you never open the app, the next time you do, once it’s more than 10 minutes old). This is the only time an image ever leaves your device, and only when you pick one and ask for replies.

The Suggest box in the app and the Contact us form on the Support page are how you reach the Murmur team. They’re one of only two places where something you write is sent to us and kept in a form we can read — and only when you press Send.

We receive exactly what you wrote, your email address (optional in the Suggest box; needed on the Contact us form, so we can answer), and a few technical details that help us fix things: whether you used the web app, the iPhone app, the Mac app or the Windows app, the app’s version, and which screen you sent it from. Nothing else — not your history, Dictionary, notes, or anything else in Murmur. If you dictate your message, it goes through the same speech-to-text and cleanup as any dictation first.

Messages are stored privately in cloud storage through our hosting provider, Vercel, where only the Murmur team can read them. The team also gets an email alert with a copy of your message (and your email, so we can reply), sent through our email delivery service, Resend. Resend keeps each email, including its content and delivery logs, for up to 30 days and then deletes it, under its own policy. We use what you send only to improve Murmur and to answer you. It’s never sold or shared. We keep it until it’s no longer needed or you ask us to delete it.

“Help Murmur understand voices like yours” in Settings is off unless you turn it on. When it’s on, and only after you correct a dictation, Murmur asks: “Send this recording to help Murmur learn voices like yours?” Nothing is sent unless you press Send. “Don’t ask again” turns it off. While it waits for your answer, the sample stays only in the app’s memory (the only copy in storage is your last recording, kept on your device for recovery). On iPhone, it never asks while “Keep my voice on this iPhone” is on — your voice isn’t recorded then.

What’s kept: that recording, what speech-to-text heard, what Murmur wrote, your corrected text, the app and version you used, and — only if you add it — how you describe your voice (for example, an accent or how you speak). It’s filed under a random ID made on your device, which isn’t linked to your name, your other data, or anything else.

For how long: 12 months — then it’s deleted automatically. Delete it sooner: Settings → “Delete all my samples” deletes every sample sent from that device, right away. Turning the switch off stops new ones but doesn’t delete what you sent — use the delete button for that. (Clearing Murmur’s data on your device keeps that button working.)

Who can access it: only the Murmur team. Samples are stored privately in cloud storage through Vercel, are never sold or shared, and are never sent to an AI provider that keeps data. They’re used only to measure and improve how accurately Murmur understands voices. By sending one, you let us use it for that purpose (see the Terms).

There are no accounts, so to bring your name, Dictionary, snippets and settings to another device, Murmur gives you a one-time code: Settings → “Move my words to another device”. You choose what goes — your name, Dictionary, snippets, the people you text and your style settings are ticked; training progress, history and Scratchpad only if you tick them.

• Encrypted on your device. What you chose is encrypted on your device (AES-256-GCM) before it’s sent. The key comes from the last 8 characters of the code, which never leave your devices — the QR code carries them after a “#”, the part of a link browsers never send to a server.
• Murmur can’t read it. Our server receives only the encrypted copy and keeps it privately in cloud storage through Vercel, with the time it was made. It never gets the key, never decrypts it and never logs it.
• Deleted on first use. The first device to open the code gets the encrypted copy, and it’s deleted from our server right then. Unopened, it can’t be opened after 10 minutes, and it’s deleted within a day.
• Only the code holder can open it. Anyone with the code can open it once, so keep it to yourself. To stop people guessing codes, our server limits how often each IP address can make or open them (see Hosting, logs and cookies).

On the new device, everything is decrypted on that device and added to what’s already there — nothing is removed.

Murmur asks only for what it needs, and only for these reasons:

Microphone

To hear your dictation — and, when you transcribe a meeting, the room or your side of a call. Every platform.

System Audio Recording Only (Mac)

Only for transcribing a meeting on a call, and Transcribe what’s playing: to hear your computer’s sound (the call, or what’s playing) while it records. macOS asks the first time you start; change it anytime in System Settings › Privacy & Security › Screen & System Audio Recording. (Windows has no permission for this.)

Speech recognition

iPhone and Mac: to show your words live while you talk (Apple’s speech recognition), and on iPhone to do all the work when “Keep my voice on this iPhone” is on.

Keyboard “Allow Full Access” (iPhone)

iPhone keyboards can’t use the microphone, so the Murmur keyboard opens the Murmur app to listen, and Full Access lets the two share your Dictionary, settings and the dictation to type. It also lets the keyboard’s tone buttons and voice edits reach Murmur’s server. It isn’t used to log keystrokes or read your clipboard, and nothing you type is sent — except the text a voice edit or tone button changes, when you use one.

Notifications (iPhone, optional)

For an occasional reminder to dictate. Never includes your words.

Reminders and Calendar (iPhone, optional)

Only to add the reminders and events you ask for by voice (“Murmur, remind me to call Sam at 3”). Calendar is add-only — Murmur can’t see your calendar. Reminders: iPhone only offers full access, but Murmur only adds the ones you ask for and removes one if you tap Undo. Neither is sent anywhere.

Accessibility (Mac)

To type your words into other apps, to read text you highlight for a voice edit, and for hold-to-talk (Fn) and Esc.

Input Monitoring (Mac, if asked)

Only if macOS needs it for hold-to-talk. It lets Murmur notice its shortcut keys; it never records what you type.

Windows

UI Automation, to read text you highlight for a voice edit and to check the words Murmur typed before changing them; a keyboard hook that sees which keys are pressed so it can catch Murmur’s shortcuts (like holding Ctrl+Win) — it never records what you type; and simulated Ctrl+V to paste.

Launch at login (Mac and Windows)

On the first time you open the app, so Murmur is ready when you need it. Turn it off anytime in Settings or the Murmur menu.

We use a small number of service providers, only to run Murmur. They may use what they receive only to provide their service to us.

Vercel

Hosts the website, the app and our server; runs Vercel AI Gateway (which routes requests to the AI providers below) and the private storage for suggestions, voice samples and move codes’ encrypted bundles.

xAI

Speech-to-text, with zero data retention.

Anthropic

AI cleanup, voice edits, translation, summaries, answers about a transcript, follow-up emails and screenshot replies (Claude), with zero data retention.

Resend

Delivers the email alerts about suggestions and support questions to the team.

Apple, Google

Their own speech recognition on your device or browser shows your words live (see above), under their own privacy policies. Apple and Microsoft make the operating systems Murmur runs on; we don’t send them your data.

If you try the iPhone app through Apple’s TestFlight, Apple may share with us crash reports and any feedback you choose to send through TestFlight, under Apple’s terms.

Our AI providers (xAI for speech, Anthropic for polishing, through Vercel’s AI Gateway) are US companies, so what you dictate is processed in the United States. As described above, they don’t keep it.

To follow our providers’ rules, we check which country (and region) your connection comes from, using the location our host Vercel derives from your internet address, so we can tell whether Murmur is available there. We don’t store it.

Your data is stored and processed in the United States.

Like any website, our hosting provider (Vercel) keeps short-lived, standard request logs — things like the time, which page or service was asked for, and technical details such as your IP address and browser type. Our server’s own error logs never include your audio or anything you said or typed — not even when something goes wrong.

To stop abuse, our server briefly counts requests from each IP address over the last few minutes. That count lives only in the server’s short-term memory and is never saved.

No analytics, ads or tracking. Murmur has no analytics, tracking pixels, ad networks or advertising IDs, and we never sell or share your data. No cookies are set for you on this website or in the apps. The only cookie Murmur uses is a sign-in cookie for the team’s private dashboard, set only for team members who sign in to it.

Your recordings and words

Not kept by us or by our AI providers (zero data retention).

Your last recording

On your device only, for up to 24 hours. Each new dictation replaces it; “Delete it now” deletes it sooner.

Screenshots for replies

Not kept.

Files you transcribe

Not kept — not by us, not by our AI providers, and not for “Recover last dictation”. Only the transcript stays, in History on your device, until you delete it.

Meetings (Transcribe and note a meeting)

The sound: not kept — not by us, not by our AI providers, and not on your device. The transcript and notes: in History on your device, until you delete them.

Questions, answers and follow-up emails

Not kept by us or by our AI providers. They stay with the transcript, in History on your device, until you delete it.

What you save in the app

On your device, until you delete it.

Suggestions and support questions

Until no longer needed, or until you ask us to delete them.

Email alerts at Resend

Up to 30 days.

Voice samples (if you send them)

12 months, or until you delete them.

Move codes (encrypted)

Deleted when opened. Unopened ones stop working after 10 minutes and are deleted within a day.

Mac and Windows troubleshooting log

On your computer only; entries older than 14 days are cleared out whenever Murmur runs.

Hosting request logs

A short time, then they expire on their own.

Murmur isn’t for children: you must be 18 or older to use it (see the Terms), and we don’t knowingly collect personal information from anyone under 18. If you believe a child has sent us something (for example, a suggestion with their email), tell us and we’ll delete it.

• Delete your voice samples: Settings → “Delete all my samples”, anytime, on the device that sent them.
• Delete a suggestion or support question, or ask anything: use the Contact us form or the Suggest box in the app (add your email so we can reply). Tell us roughly when you sent it and the email you used, if any, so we can find it.
• Delete your last recording: “Delete it now”, beside “Recover last dictation”.
• Everything else is on your device, where you can delete it yourself: Settings → Clear all local data, or delete the app.

Because there are no accounts, we hold nothing else that’s tied to you.

The Murmur keyboard’s next-word suggestions use word counts from English sentences shared by the Tatoeba project, under the CC BY 2.0 FR license.

If what Murmur does with your data changes — for example, when accounts arrive — we’ll update this page first and change the effective date at the top. See also our Terms of use.