Skip to content
Respond
DesktopiPhoneWhy RespondPlans
Your accountGet Respond
DesktopiPhoneWhy RespondPlansYour account
Privacy PolicyTerms of ServiceRefund Policy

Your privacy.
In plain words.

Last updated 3 October 2026

On this page
01Who we are02What Respond reads from your device03Respond for iPhone04Where your highlighted text goes05What we store06Service providers07Analytics and tracking08How long we keep things09Your rights10International transfers11Children12Changes to this policy13Contact
A question? Say hello

Respond is a desktop application that translates text you highlight in other applications. That means it reads text from your screen, and you are entitled to know exactly how much, where it goes, and what is kept. This policy answers those questions in plain terms.

The short version: Respond reads only the text you deliberately highlight, hears you only while you ask Ondy or use Live translate, and looks at your screen only when you ask Ondy about it. That text is sent to the translation service you have chosen so it can be translated, and it is not stored by us afterwards.

There is one place where that last sentence does not hold, and it is the most important thing on this page. When you share a Chat Mode conversation with a second phone, both sides of it sit on our server for as long as the room is open. What that means is set out below.

Desktop microphone, screen and captions

On Windows and Mac, Respond uses your microphone only when you start it: while you ask Ondy, or, on Windows, while Live translate is on. Screen Captions captures system audio only after you choose a screen or an app and start a session. System audio may include other audible apps; the captions feature does not use your microphone. On a Mac, macOS asks you to allow Screen Recording before Respond can hear that sound.

When you ask Ondy about your screen, by holding the Ask Ondy shortcut or by tapping it and dragging over an area, Respond takes one picture of that screen or that area and sends it with your question to the AI provider selected by your plan or settings, through the Respond service for customer editions or directly using your configured key in Respond Internal. The picture is used only to answer that question, and the Respond service does not keep it. Respond keeps a copy with that chat on your computer, encrypted, so you can see it when you open the chat again, and deletes it when the chat is deleted. Pictures you attach from a file are kept the same way. On a Mac this also needs Screen Recording.

Audio clips are sent to OpenAI for transcription, through the Respond service for customer editions or directly using your configured key in Respond Internal. Captions can also send the resulting text to your chosen translation provider. No screen video is saved or uploaded. Audio is held temporarily to finish transcription or retry a failed section, and discarded when you close the session or quit the app.

Live translate (Windows)

Live translate starts only when you right-click a text box, choose it, choose your languages and press Start. While it is on, each phrase you say is sent to our service. We pass the audio to OpenAI to write it down, and the words to your translation engine. Respond then types the translation into the box you chose, as if from a keyboard.

It types only into that box's window, and only while that window is in front: a click into another window stops it. It never presses Enter, so nothing is sent until you send it yourself. Redo takes back only what Live translate typed. It never reads or uses your clipboard. The audio is held only for that request and is not saved. Pause stops listening, and Stop or the end of the session lets the microphone go.

Completed transcripts, translations and summaries remain in encrypted history on your device until you delete them in Settings → Captions. Audio is not saved in history. Saving exports a plain text file to the location you choose; Copy uses your clipboard. Deleting a history item does not remove copies you exported. Summarize sends transcript text to the AI provider selected by your plan or settings.

Who we are

Respond is operated by the developer trading as Respond ("we", "us"). You can reach us at info@withrespond.com for any question about this policy, or to exercise any of the rights described below.

What Respond reads from your device

Text you highlight

When you highlight text and activate Respond, the application reads that selection through the Windows accessibility system. It never copies it and never touches your clipboard.

On a Mac, Respond reads the selection only through macOS Accessibility, the permission you allow when you set it up, and never by copying. WhatsApp for Mac does not share what is highlighted inside a message, so there Respond reads the whole message under the pointer instead.

Respond does not record keystrokes, and does not monitor or read anything you have not highlighted. The only picture of your screen it takes is the one you ask Ondy for, described above. The only keys it ever presses are the letters of a translation, typed into the box you chose for Live translate.

What Respond refuses to read

  • Any control that the operating system identifies as a password field. The selection is discarded before it leaves the application.
  • Any application on your block list. Respond ships with password managers already blocked, including 1Password, Bitwarden, KeePass, Dashlane, LastPass, NordPass, RoboForm and Enpass, plus the Windows credential and elevation prompts and, on a Mac, Keychain Access, the Passwords app and the macOS password prompts. You can add to or change this list in Settings.
  • For the automatic hover preview only, text that has the shape of a credential: API keys, private keys, payment card numbers and short numeric codes. These are not previewed automatically, though you can still translate them deliberately.

Global input hook

To notice that you have highlighted something, Respond observes mouse and keyboard events system-wide. This is used solely to detect selection gestures, your chosen shortcut, and, while Live translate is on, a click into another window, so it can stop. Key presses are not recorded, stored or transmitted.

Respond for iPhone

The iPhone app is a separate application with its own permissions. It never reads your screen and has no equivalent of the desktop selection feature. It only ever sees what you type into it, speak into it, point its camera at, or attach to a question for Ondy; and, through the Respond keyboard, what you have typed in the field it is attached to, and a message you copied when you press its Respond key.

Asking before anything leaves your iPhone

The first time something you give the app would go to one of the companies that translate, read aloud and answer for it, the app asks you first. It uses a standard iPhone alert that says what is sent (text, speech, photos and documents) and names who it goes to: Google, OpenAI, Anthropic and ElevenLabs. Until you tap Allow, nothing goes to any of them. Anything your iPhone can do by itself, such as translating a language pair it has on the device, carries on as normal.

If you tap Don't Allow, anything that needs them stops and says why. You can change your answer at any time in the app's Settings, under Privacy & your data, AI services. The Respond keyboard follows the same answer and never asks on its own. Everything below about what goes where applies only once you have allowed it.

The microphone

Used only while you are talking to the app: dictating on the home screen, holding the talk button in Chat Mode, asking Ondy a question out loud, practising how to say a phrase, and saying the phrase onboarding asks you to say. Not at any other time. Speech is turned into text by Apple's speech recognition, and where that happens depends on whether you have a connection:

  • With a connection, the audio goes to Apple to be recognised — the same service the iPhone's own dictation uses, under Apple's privacy policy, not ours. This is the default because it is markedly more accurate, and because Respond listens for both of your chosen languages at once and needs Apple's own measure of which one it heard. That measure is only available this way.
  • Without one, recognition happens on the iPhone itself, for the languages it can manage offline.

Either way, no audio is ever sent to us, none is recorded, and none is kept — by us or, for this purpose, by anyone. Only the finished text goes on to be translated.

Two exceptions, and both are the same set of languages. The iPhone has speech recognition for sixty-three languages; Respond translates a hundred and thirty-six. For the rest — Shona, Zulu, Swahili, Xhosa, Amharic, Filipino and some seventy others — there is no recogniser on the phone to use, and holding the talk button would otherwise produce nonsense. In those languages the recording is sent to our own service, which passes it to OpenAI to be written down and sends the words back. The recording is held in memory for the length of that one request, is never written to disk or to a database by us, and is never logged. OpenAI states that data sent through its API is not used to train its models. The same goes for a dialect the iPhone can only hear as its parent language, such as Sicilian heard as Italian, because that recogniser can only write the parent language's words.

  • With Respond Pro or Respond Max. Dictating on the home screen, holding the talk button in Chat Mode, asking Ondy out loud or practising a phrase in one of those languages or dialects records what you say and sends it the way just described. Every other language still goes to Apple.
  • Setting the app up, with no account and no payment. Onboarding asks you to say a phrase out loud and marks it word by word. In the languages the iPhone can hear, that happens on the phone and nothing leaves it. In the others, once you have allowed it, that one recording of your voice goes to OpenAI by the same route — before you have an account, before you have paid for anything, and whether or not you ever do. It is handled the same way: held for the length of the request, not stored, not logged.

This page used to say the OpenAI route was reached only with Respond Pro. That was true of Chat Mode and never true of onboarding, and the correction matters most to exactly the people this app is for: it is the languages the iPhone cannot hear that take this path, so the first recording of a Shona or Amharic speaker's voice left the phone while the policy said it could not.

Only the finished sentence — text, not audio — is translated, and only in the same way as anything else you type.

The camera

The camera translator reads the words on your iPhone, with Apple's on-device text recognition. No photograph or video is sent anywhere; only the recognised text is translated. If you choose to save a photo with its translation, it goes into your own photo library and nowhere else.

The camera is also used when you take a photo to attach to a question for Ondy. That photo is sent, as the next section describes.

Ask Ondy

Ondy answers questions about how to say things, and needs Respond Pro or Respond Max. What you ask, and any photo or document you attach to the question, goes to our service, which passes it to OpenAI or Anthropic, whichever engine you have chosen, and sends the answer back.

  • A photo is redrawn on your iPhone before it is sent, which leaves its location and every other piece of metadata behind. The original file is never sent.
  • For a document, only its text is sent.
  • We do not store your questions, photos or documents. They are held for as long as it takes to answer, and our logs record only how long a question was.
  • Your conversations with Ondy are kept on your iPhone, not on our servers, and you can delete any of them in the app.

Reading a translation out loud

Tapping the speaker under a sentence reads it aloud, and where the voice comes from decides whether that sentence leaves your iPhone.

  • If your iPhone has one of Apple's enhanced or premium voices installed for that language, the reading happens on the device and nothing is sent anywhere.
  • Otherwise the sentence is sent to Google's text-to-speech endpoint — the same one the speaker button in Google Translate uses — and the audio that comes back is played. A new iPhone ships only the basic voices, so this is the usual case rather than the rare one.

This deserves saying plainly, because it was missing from this page altogether and because it is easy to assume otherwise: reading aloud goes to Google whichever translation engine you have chosen, including On device. Choosing On device keeps the translating on your iPhone. It does not keep the reading there. What Google receives is the sentence being read and the language to read it in, under its own privacy policy rather than ours. If you would rather nothing went, do not use the speaker, or install one of Apple's better voices for that language in the iPhone's own Settings and the reading stays on the device. Until you allow AI services, the iPhone's own voices do all of the reading.

A dialect is read differently. With Respond Pro or Respond Max, a dialect such as Sicilian or Egyptian Arabic is read by a voice made for it: the sentence goes to our service, which passes it to OpenAI, or to ElevenLabs where that dialect has a voice of its own, and plays the audio that comes back. What they receive is the sentence and the dialect it is in.

The Respond keyboard

The keyboard asks for what iOS calls Full Access. iOS shows a strongly worded warning when you turn this on, and it is right to: a keyboard with Full Access could send everything you type somewhere. So here is exactly what this one does.

  • Full Access is needed for one reason only — without it, iOS gives a keyboard no network connection at all, so it could not translate anything. The keyboard tells you this plainly, and it still types normally without it.
  • Nothing is sent anywhere as you type. Text leaves your iPhone only when you press the Translate key, and only the text you are translating.
  • The Respond key suggests replies to a message you have copied. Pressing it is the only time the keyboard reads what you copied, and that message is sent, the same way as a translation, so replies can be written for it.
  • The keyboard does not log, store, or transmit your keystrokes. It keeps no history.
  • Spelling suggestions come from your iPhone's own dictionary and are worked out on the device.

Respond Pro and Respond Max on iPhone

Respond Pro and Respond Max are bought through Apple. The purchase itself tells us nothing about you: Apple hands over neither your name, nor your email address, nor your payment details. What our service receives is the receipt Apple signs for the purchase, which identifies the subscription and when it expires and nothing else. To check that a subscription is still active, our service asks Apple's App Store Server API about that receipt's transaction, and Apple answers with its status and nothing about you.

That is true of the purchase and only of the purchase. It is not a statement about the app as a whole, and an earlier version of this page read as though it were. If you make a Respond account, we do hold your name and your email address, along with the other answers you gave when you set the app up. What we store, below, lists all of it.

We record how many characters that subscription has translated in the current month, so that the included allowance can be enforced. That count is kept against Apple's subscription identifier, which is not linked to any name, email address or device.

Conversations

Chat Mode keeps your past conversations on your iPhone so you can look them up again. That history is stored on the device only, is never uploaded, and is deleted when you delete it or when you delete the app.

Sharing a conversation with a second phone

This is the exception to the paragraph above, and it was not on this page at all. Chat Mode can put a square on your screen for the other person to scan, or give you a six-letter code to read out. When it does, Respond opens a room on our server, and that room is how the two phones pass words between them.

While the room is open, both sides of the conversation are on our server — every line either of you speaks or types, and the translation of each one. There is no way around it: that is the mechanism by which what you say reaches the other phone at all. So for as long as a room is open, the sentence "conversations are stored on your device only and are never uploaded" is not true of that conversation.

Being exact about what that means:

  • The lines are held in our server's memory. They are not written to a disk, not written to a database, and not backed up. Restarting our service loses every room open at the time, which is by design rather than by accident.
  • A room is dropped once ninety minutes have passed with neither phone contacting it, and everything in it goes at that moment. The clock runs from the last contact rather than from when you started, so a conversation left open on screen keeps its room alive.
  • The code is the only key. Anyone holding it can open the page and read the conversation while the room lives — a person who photographed the square over your shoulder, or anyone the guest forwards the link to. There is no sign-in on that page and no second check. Treat the square the way you would treat the screen itself.
  • Ending the conversation on your phone stops anything further being added to it. The lines already there stay readable until the room is dropped.
  • We do not read rooms, and nothing in one is kept, counted or attached to your account. Our server does not translate them either: your phone does the translating and posts the result back, which is why a guest gets the same quality you do.

The practical advice, and we would rather give it than dress this up: a room is the right tool for a conversation at a counter, and the wrong one for anything you would not want a third person to read. For those, hand over the phone, or type into the app yourself.

If somebody has asked you to join a conversation

You need no app, no account and no sign-in to join one — you open a link or type a code, and the conversation appears in your own language. That page asks you for nothing: not your name, not your email address, not your phone number. It sets no cookie and runs no tracker.

What it does do is worth knowing before you use it. Everything you send goes into the room described above, on the same terms, readable by anyone holding the code and kept until the room is dropped. The translating is done by the phone of the person who invited you, using whichever service they have chosen, so your words pass through their app on the way. And if you use the microphone on that page, the recognition is your own browser's rather than ours — some browsers do that on the device and others send the audio to the company that makes the browser, which is a matter between you and them and not something we can see or control. Typing instead avoids it entirely.

Where your highlighted text goes

Translation requires sending the highlighted text to a translation service. Which service depends on the engine you select in Settings.

Engine Who receives the text When it is used
Google Google's public web translation endpoint The default free engine, used unless you choose otherwise
Claude Our translation service, which passes it to Anthropic Respond Pro and Respond Max subscribers who select Claude
GPT Our translation service, which passes it to OpenAI Respond Pro and Respond Max subscribers who select GPT
DeepL DeepL, directly from your computer using your own key Only in the internally distributed edition

Interface labels are also translated through the Google engine when you set Respond to an interface language other than English. These are fixed strings from the application, never your own text.

We do not store the text you translate. Our service passes it to the provider, returns the translation to your computer, and retains no copy of either. The providers listed above process it under their own terms and privacy policies. The single exception is a Chat Mode room, where both sides of the conversation are held on our server while the room is open.

What we store

If you never create an account

Nothing. The free version requires no account and no sign-in. Your settings live in a file on your own computer and are never sent to us.

If you make a Respond account

We store, on our servers:

  • Your email address, from the account you sign in with.
  • Your name, as you typed it when the app asked for it. It is kept on the account rather than on the phone so that it follows you to your next one.
  • The three answers the app asks for while setting itself up: the language you already speak, the language you are learning, and your answer to why you are here. They are stored against the same account record as your name and email address, which means they are not anonymous and it would be wrong to imply they were. This page previously listed none of them.
  • Your subscription status and billing period, along with the identifiers our payment processor uses to reference your subscription.
  • A monthly count of characters translated and translations performed. This is a running total used to enforce the fair-use limit. It is a number, not content: we cannot reconstruct anything you translated from it.

On your iPhone

Your settings, saved phrases, conversations, Your words and your conversations with Ondy are kept in the app's own storage on your iPhone and are not sent to us. Deleting the app deletes them.

On your own computer

Your settings are stored in your user profile on your computer. Any API keys you enter yourself are encrypted, using the Windows Data Protection API on Windows or a key kept in your keychain on a Mac, tied to your account on that computer, and are never sent to us or exposed to the application's interface after you save them.

Service providers

We use the following providers. Each receives only what it needs to do its job.

Provider Purpose What it receives
Anthropic, OpenAI Translation for Respond Pro and Respond Max, Ask Ondy, suggested replies, second opinions on a translation, and pronunciation guides The text being translated or asked about, and any photo or document text you attach to a question for Ondy
OpenAI Writing down speech in the languages and dialects the iPhone cannot hear: with Respond Pro or Respond Max, and while setting the app up, with no account The recording of that one turn, for the length of the request
OpenAI Writing down speech on Windows and Mac: asking Ondy out loud, live captions, and Live translate on Windows The audio of that phrase, for the length of the request
OpenAI, ElevenLabs Reading a dialect aloud, with Respond Pro or Respond Max The sentence being read and the dialect it is in
Apple Sign in with Apple, App Store subscriptions, and speech recognition on iPhone Your name and email address, or a private relay address, if you sign in with Apple; the signed receipt for a subscription; speech sent for recognition, under Apple's own privacy policy
Google Free translation engine, reading translations aloud, and sign-in The text being translated; the sentence being read aloud, whichever engine you chose; your email address on sign-in
Supabase Accounts and the subscription database Your name, email address, the languages you set, why you said you are here, and your subscription status
Stripe Payment processing Your payment details, which we never see or hold
Render Hosting for our translation service and for Chat Mode rooms Requests in transit, including your IP address; the contents of an open room, held in that machine's memory
GitHub Distributing application updates Your IP address when checking for or downloading an update

Analytics and tracking

Respond contains no analytics, no usage tracking and no advertising identifiers. It does not record what you translate, how often you use it, or which applications you use it in. This website sets no cookies and runs no third-party trackers.

Crash reports

The one exception is crash reporting, and it is off unless you turn it on in Settings. With it off, nothing is ever sent. With it on, a crash sends us:

  • The error message and the stack trace showing where in our code it happened.
  • The Respond version, the edition, and your Windows or macOS version.
  • A random identifier that is regenerated every time Respond starts and never written to disk. It exists only to group several errors from one crash together, and cannot be used to recognise you or your computer on a later run.

Before a report leaves your computer, your account name on that computer is removed from any file paths it contains, and anything shaped like an API key, token or password is replaced. The same removal runs again on our server. Reports never contain text you translated, translations, your settings, or your API keys.

Reports go to our own service. They are not sent to any third-party crash reporting or analytics company. You can turn reporting off again at any time in Settings, and it takes effect immediately.

Reports are deleted after 90 days. That deletion runs by itself, once a day, against our database — it is not a note in a calendar and it does not wait for anybody to remember it.

How long we keep things

  • Translated text: not retained, with the one exception of a Chat Mode room, which is the last line in this list.
  • Audio from asking Ondy, live captions and Live translate: not kept. It is held only while it is written down.
  • Monthly usage counts: retained while your account exists, so limits can be applied across a billing period.
  • Account and subscription records: retained while your account exists, and after closure only where we must keep transaction records to meet tax and accounting obligations.
  • Server logs: kept briefly for security and debugging. They record errors and request metadata, not translated content.
  • Crash reports, if you switched them on: 90 days, deleted automatically once a day.
  • Chat Mode rooms: held in memory only, and dropped once ninety minutes have passed without either phone contacting the room. Nothing about a room survives it.

Your rights

Depending on where you live, you may have the right to access, correct, export or delete the personal information we hold, to object to or restrict its processing, and to complain to your local data protection authority. Email info@withrespond.com and we will action your request. In the iPhone app you can delete your account yourself, under Settings, Privacy & your data, Account controls. If you made it with Sign in with Apple, the app asks Apple to confirm, and we then revoke Respond's sign-in with your Apple ID, so Respond no longer appears under Sign in with Apple in your Apple settings. To delete your account entirely, ask us and we will remove your account record — name, email address, the languages you set and the answer you gave about why you are here — along with your usage history.

International transfers

Our providers operate internationally, so your information may be processed outside the country you live in, including in the United States. Where required, transfers rely on the safeguards those providers have in place, such as standard contractual clauses.

Children

Respond is not directed at children under 13, and we do not knowingly collect their personal information.

Changes to this policy

If this policy changes in a way that materially affects how your information is handled, we will update the date at the top and, for account holders, notify you by email before the change takes effect.

What changed on 3 October 2026

On Windows and Mac, a picture you send to Ondy with a question now stays with that chat on your computer, encrypted, so you can see it again when you open the chat. It is deleted when the chat is deleted, and the Respond service still does not keep it. This page used to say the picture is not saved.

What changed on 2 October 2026

Live translate is new on Windows. It listens only while you have it on, sends what you say to be written down and translated, and types the translation into the box you chose. This page now says so, names OpenAI for writing down speech on the desktop, and says that audio is not kept.

This page also used to say that on Windows Respond sometimes read a selection by briefly copying it. It does not: it reads only through Windows accessibility and never touches your clipboard.

What changed on 1 October 2026

The iPhone app now asks before anything you give it goes to Google, OpenAI, Anthropic or ElevenLabs, and this page says so, along with where to change the answer.

It also catches up with the iPhone app. This page used to say the microphone was only used in Chat Mode and while setting the app up, that no photo was ever sent, and that the keyboard only sent text when you pressed Translate. The microphone is also used for dictation, for asking Ondy out loud and for practising a phrase; a photo you attach to a question for Ondy is sent to answer it; and the keyboard's Respond key sends the message you copied. Ask Ondy, dialect voices (including ElevenLabs, which this page did not name before), subscription checks with Apple, Sign in with Apple, and deleting your account in the app are now described too.

Respond is now on the Mac as well. This page names the Mac wherever it named Windows, and says what the Mac's permissions are for: Accessibility to read what you highlight, Screen Recording for Ondy's picture of your screen and for the sound live captions hear, and the microphone for Ask Ondy.

It also catches up with Ask Ondy, which can look at your screen when you ask it to. This page used to say Respond never takes a picture of your screen. It now says when it does, where the picture goes, and that it is not kept.

What changed on 2 September 2026

A second subscription level, Respond Max, went on sale inside the iPhone app. It reaches the same two translation providers as Respond Pro, on their larger models, so nothing about where your words go or who receives them has changed. This page names it everywhere it already named Pro, so that no row of the table above reads as though it covered only one of them.

What changed on 18 August 2026

The app moved and this page did not. None of the following is a change to what Respond does — it is this page catching up with what it has been doing, and the gaps were ours.

  • Chat Mode rooms were not mentioned at all, while this page said conversations never leave your device. They are now described in full, including that anyone holding the code can read one.
  • Sending speech to OpenAI was described as something only Respond Pro does. It also happens while you set the app up, with no account and no payment.
  • Reading a translation aloud was missing entirely. It sends the sentence to Google whatever engine you have chosen.
  • Your name, the languages you set and your answer to why you are here were stored but not listed, and a line about Apple's payment details read as though we held no name at all.
  • Deleting crash reports after 90 days was stated as a promise before anything ran it on a schedule. It runs daily now, so the promise stands as written.

Contact

Questions, requests or complaints: info@withrespond.com.

Respond© 2026 Respond.
PrivacyTermsRefundsAccountSay hello ↗