Privacy

Every dictation is turned into text in the cloud, so LotusQ needs an internet connection. With push-to-talk, your audio streams from the moment you press your key until you let go. In hands-free mode, your computer listens for speech and sends audio only once it hears you, starting about 0.3 seconds before you speak. LotusQ doesn't keep your recording, except briefly in memory so you can retry a failed dictation (section 6). Our speech processors may keep audio for a limited period under their own terms, as described in section 2 below.

The desktop app uses a lightweight native client and turns speech into text in the cloud, so you do not need to download, manage, or update a transcription model locally.

LotusQ never saves audio recordings, not on your computer and not on our servers (the retry exceptions below are in memory only). Your audio is held in memory only while a dictation is in progress, then dropped. There are two exceptions, so you can retry without speaking again. If a dictation fails, LotusQ keeps the recording in memory only, for up to 15 seconds, so you can try again. It is never saved to disk. And if a Compose request fails, its recording stays in memory for up to 15 seconds while the "try again" panel is on screen.

LotusQ does not include any analytics or crash-reporting tools, and the app's own log file stays on your computer. Our website, lotusq.io, counts visits with Pirsch Analytics, a cookieless tool whose servers are in Germany. It sets no cookies and stores nothing on your device. It never stores your IP address: it turns your IP address and browser details into an anonymous code that changes every 24 hours, so it cannot recognise you on another day or on another site. We see only totals, such as how many visits came from a particular link. The LotusQ app itself contains no analytics. It sends your audio to our transcription API while you dictate, and, when you use Compose, your instruction, selected text, app name, your name for sign-offs (unless you clear it) and your saved snippets (except ones you exclude), plus nearby on-screen text if you turn that on, to our API. When you use Translate (from version 2.10.0), it sends your audio, as for a dictation, or the text you selected, with the language you chose, to our API; only text, never the audio, goes on to our AI text provider (section 3). It also talks to our servers to sign you in, sync your dictionary and snippets, check your plan, report setup progress, report how each Translate press went (content-free, from version 2.10.0; section 1) and check for updates. When LotusQ doesn't recognise the app you're dictating into, or knows it only as a browser, it sends the app's name (and, if you turn it on, the window's title) so a third-party AI model provider can say what kind of app it is (section 5), and, for some accounts, it sends the auto-learn corrections it finds (section 4). Section 1 below lists everything that leaves your computer.

Everything the app sends, including your audio, your text and your account data, travels over an encrypted connection. LotusQ does not keep the audio after the transcript is returned (see section 6 for retries).

So you can find and reuse your words, LotusQ keeps the text of up to your last 10,000 dictations, your dictionary, your snippets and the corrections it has learned on your computer. Signing out does not remove them. Section 6 below has the details.

Exactly what happens to your audio and account data.

We collect what we need to run the service: your email address (to sign you in and to send you the emails described in section 11), the name you give when you sign up, your subscription status, workspace data, your dictionary and snippets, usage counts (such as words dictated and seconds of audio, never the text itself), daily totals of accounts and usage, which contain no information about any one person, and any messages you send us. When you first set up the app, you can tell us where you heard about LotusQ by picking from a short list. We store your answer with your account, and use it only to count which places bring people to LotusQ. You can also tell us what kind of work you do. It is optional. We store your answer with your account and use it only to count which kinds of work LotusQ is used for. You can change it in Settings at any time; to have it removed, contact us or delete your account. From version 2.10.0, when you use Translate, the app records content-free usage details for each press: the target language, whether you spoke or translated selected text, whether it worked (and if not, the kind of error), how long it took, and the app version and operating system. These details are linked to your account so we can leave our own test accounts out of the figures. We keep them for up to 90 days to keep Translate reliable, then delete them. They never include what you said, the text you selected or the translation. We do not keep your audio recordings, apart from the brief in-memory hold for a retry described in section 6. LotusQ does not include analytics or crash-reporting tools, and the app's own log file stays on your computer and is never uploaded. Our website, lotusq.io, counts visits with Pirsch Analytics, a cookieless tool whose servers are in Germany. It sets no cookies and stores nothing on your device. It never stores your IP address: it turns your IP address and browser details into an anonymous code that changes every 24 hours, so it cannot recognise you on another day or on another site. We see only totals, such as how many visits came from a particular link. The LotusQ app itself contains no analytics. Our servers keep logs to run and debug the service. These logs do not contain your audio, but they are not anonymous: some include your account ID, and some include your email address. To stop the same person taking the free trial more than once, we also keep a record of who has already had a trial. This record holds a keyed, one-way hash (a cryptographic fingerprint made with a secret key) of your email address, after normalising it (for example, ignoring dots and "+" tags in Gmail addresses), and a similar keyed hash of your Google account ID if you sign in with Google. Your email address itself is not stored in this record. We use it only to prevent abuse of free trials.

Here is everything the app sends from your computer:

• Your audio, while you dictate, with a dictation ID, your platform, the app version and your time zone (section 2).
• The words in your dictionary, which may be sent along with your audio as a vocabulary hint to help recognise words specific to you, and the name of the app you are dictating into, used to tidy the text.
• Your sign-in details, to our authentication provider, and requests for your profile, plan and billing.
• Your dictionary, snippets and learned dictionary words, synced to your account whenever you are signed in.
• Your computer's platform (Windows, macOS or Linux), reported when you sign in, and your progress through setup. On a Mac, setup reports also say which permission steps are done, with the app version, the macOS version and the chip type.
• When you use Compose, the details in section 3, plus any rating or example you choose to share.
• The name of an app LotusQ doesn't recognise or knows only as a browser, the website's address if LotusQ can tell which site it is, and, only if you turn it on, that window's title with email addresses, links and numbers removed, to pick a writing style (section 5).
• For some accounts, auto-learn corrections LotusQ finds, with the text of that field, except ones that would override a word you added to your dictionary yourself or undo one already in it (section 4).
• From version 2.10.0, when you use Translate, your audio, turned into text on our servers as for a dictation, or the text you selected (up to 5,000 characters), with the language you chose; only text, never the audio, goes on to our AI text provider (section 3).
• From version 2.10.0, the content-free details of each Translate press described above.
• A periodic check against our release feed for updates.
• Anything you send us yourself, such as a support request or feedback.

There is no speech model on your computer, so every dictation needs an internet connection. With push-to-talk, your audio starts streaming the moment you press your key and stops when you let go, so even a very short or silent press can send a moment of audio, which LotusQ then throws away. In hands-free mode, the microphone stays open and your computer listens for speech; audio is sent only once it hears you, starting about 0.3 seconds before you speak. Your audio travels over an encrypted connection to LotusQ's servers, which pass it to one of our speech processors to turn it into text. Your dictionary words may be sent along with your audio, as a vocabulary hint, to help recognise words specific to you. LotusQ never writes your audio to disk, on your computer or on our servers: it is held in memory only while the dictation is in progress (and re-sent once if the first attempt fails), then dropped, unless you can retry a failed dictation (section 6). LotusQ never uses your audio to train models, and our speech processors do not use it to train models either. Our speech processors handle audio under their own terms, and may process it in countries other than your own. The usage policy for the interface we call says one of our speech processors may keep prompts, contextual information and output — for us, your audio and its transcript — for up to 55 days to detect and prevent abuse, and that its staff may review content its systems flag. We have been granted an exception to that processor's prompt-logging policy, so the prompts we send it are not stored for abuse monitoring. That exception is written about prompts, and against that processor's cloud abuse-monitoring policy, while the usage policy for the interface we actually call covers a wider set, states the 55-day period and describes no exception, so we tell you both rather than give you only the shorter answer. Its automated safety checks still run on what we send while it is being transcribed. The exception covers only that processor; our other speech processor handles audio under its own terms. When you retry a dictation or Compose instruction whose transcription failed, LotusQ sends the same recording to our speech processors again, so a speech processor that keeps audio may hold more than one copy of it. Try again re-sends a failed dictation at most once, and Compose's retry re-sends a failed instruction's recording at most once.

Compose never streams: your spoken instruction stays on your computer until LotusQ has checked that you are not in a password field, and is discarded on your computer if you are. It is then turned into text by our speech processors in the same way as a dictation (section 2). If that fails, LotusQ keeps the recording in memory (never on disk) for up to 15 seconds while the "try again" panel is on screen, so you can retry once, and drops it sooner if the panel closes. When you use Compose, your spoken instruction, the text you selected, the name of the app you're using, your name for sign-offs (unless you clear it) and your saved snippets (except ones you exclude) are sent to LotusQ's AI text provider solely to write the result. If you turn on On-screen reading, the visible text around your cursor and, in browsers on macOS, the website's domain are sent too. This data travels over TLS, and LotusQ doesn't keep a copy of what Compose sends. Our AI text provider doesn't keep this data after writing the result, except where the law requires, and doesn't use it to train models. Our server logs for Compose hold metadata only (such as sizes, timing and errors, and in some cases your account ID), not your instruction or text, except that an unexpected server error may record part of a request in our error log. These logs are kept for a limited period. If you give a thumbs up or down on a Compose result, we keep the rating with anonymous details (such as which job Compose did and how long it took) for 12 months. If you choose to share an example, we also keep that instruction, the text Compose used and the result for 90 days to help improve Compose, then delete it; we remove personal details before using it to test future versions. To remove personal details, a shared example is processed by the same AI text provider that writes Compose results. When you speak with Translate, your audio is turned into text by our speech processors in the same way as a dictation (section 2). Only that text, never the audio, then goes to LotusQ's AI text provider with the language you chose, solely to translate it. When you translate selected text instead, LotusQ reads only the text you selected, up to 5,000 characters, and sends it with the language you chose; if the selection is longer, nothing is sent. Translate never sends the name of the app you're in, the text around your cursor or your snippets. Before Translate listens or reads a selection, LotusQ checks that you are not in a password field where your computer allows it; on Linux it can't tell, and relies on apps not letting a password be copied. If a spoken Translate fails, the recording is not kept for a retry. LotusQ doesn't keep a copy of what Translate sends, and our server logs for Translate record codes and timings, not your words. Our AI text provider doesn't keep this data after translating it, except where the law requires, and doesn't use it to train models. On your computer, your history keeps each translation and, for spoken Translate, the words LotusQ heard; for selected text it keeps the translation but not the text you selected.

Auto-learn works on Windows and macOS. About 10 seconds after LotusQ types your text, or sooner if you dictate again, it reads that text field again, on your computer, to spot any changes you made. It does not read back text that Compose inserts. It does this with an accessibility feature: Windows UI Automation on Windows, and the macOS Accessibility API on macOS. LotusQ never reads a password field. This comparison happens on your computer. For most accounts, LotusQ decides on your computer, with its own rules, whether a correction belongs in your dictionary, and the text it reads is not sent anywhere. For some accounts, LotusQ instead sends every correction it finds to our servers, which ask a third-party AI model provider whether to add it to your dictionary. That includes corrections LotusQ's own rules would turn down, such as a change to an everyday word. Only two kinds are always settled on your computer and never sent: a correction that would override a word you added to your dictionary yourself, and one that would undo a word already in your dictionary. For the others, the request carries the word or words LotusQ got wrong, your correction, the text LotusQ typed for that dictation, the text in that field after your edit, whether the opposite correction is already in your dictionary, and how many times you have made this correction before. The field can hold more than your dictation, such as words you typed yourself or text that was already there; each piece of text is cut to its first 2,000 characters. Your audio and the name of the app you were in are not sent. If the request fails or takes too long, LotusQ falls back to its own rules. Our server logs for this record your account ID, your platform, the outcome, the timing, the size of the request and the scores the model gave, never the words. How long the provider may keep what we send is governed by the provider's own terms; LotusQ does not ask it to keep any of it. The corrections it finds are saved in a file on your computer. The app's log file records only counts and timings for auto-learn, such as how many corrections it found and whether one was learned, never the words themselves. A learned word that is added to your dictionary syncs to your account like any other dictionary entry. If you undo a word auto-learn added, it is removed from your account too. On Linux, LotusQ does not read text fields back after typing.

LotusQ picks a writing style, such as email, chat or coding, from the app you are dictating into, using a built-in list of apps. When LotusQ is choosing the style for you and that list doesn't recognise the app, or knows it only as a web browser, LotusQ sends the app's name to our servers, which ask a third-party AI model provider what kind of app it is. If LotusQ can tell which website you are on, for example a website installed as its own app, the site's address is sent too, but only the site's name (such as mail.example.com), never the page you are on. This happens in the background and never holds up a dictation. The answer is saved in a file on your computer, one per app, so the same app is not asked about again for up to 30 days. Old entries are not deleted from that file automatically; it holds at most 500 apps. What you dictate and your audio are never part of this. On Windows and Linux, the app has an optional setting, Use window titles to pick a better style, which is off unless you turn it on (a Mac does not read window titles). With it on, LotusQ also sends the title of that window, for the same apps only. Before the title leaves your computer, LotusQ removes email addresses, web links (a bare site name can remain), file paths, phone numbers, long numbers and codes that look like order or ticket numbers, and our servers remove them again. Other words in a title can remain, such as a person's name or the subject of an email or document. LotusQ does not store window titles: your computer keeps the cleaned-up title and its answer in memory for about 10 minutes and never writes either to disk, and our server logs for this record your account ID, the kind of app, how sure the answer was and how long it took, never the app name, the site or the title. How long the provider may keep what we send is governed by the provider's own terms; LotusQ does not ask it to keep any of it.

On our servers, account and workspace data (including your dictionary and snippets) is kept for as long as your account is active. Our database holds no text of what you dictate: usage records are counts only, apart from Compose examples you choose to share (section 3) and the content-free Translate usage details in section 1, which are deleted after 90 days. LotusQ does not keep audio (section 2), apart from two brief retry exceptions; one of our speech processors states it may keep it for up to 55 days under its own terms, and we hold an exception to that processor's prompt-logging policy. If a dictation fails, LotusQ keeps the recording in memory only, for up to 15 seconds, so you can try again. It is never saved to disk. A failed Compose instruction's recording is held the same way (section 3). On your computer, LotusQ keeps the text of up to your last 10,000 dictations (and, for Compose, your spoken instruction) with no time limit unless you choose one under Keep history for in Settings (1 day, 30 days or 1 year), your dictionary and snippets, a file of corrections you have made (the words as dictated and as you corrected them), a file of suggested dictionary words (short phrases of up to four dictated words), a file of the kind of app each app LotusQ has asked about is (its name, and the website address where there is one; an answer is reused for up to 30 days, but old entries are not deleted automatically, and the file holds at most 500 apps and can be read only by your user account on that computer) and a log file, which records neither the words auto-learn learns nor your full dictations. Signing out does not remove any of this, but you can clear your history with Clear all history in Settings. There is also no button to delete your account: to delete it, email [email protected], and we will delete your account and the data linked to it within 7 days. Some records are kept after deletion: messages you have sent us as feedback, the customer record held by our payment processor, the trial history record described below, our server logs until they expire, Translate usage details, no longer linked to you, until they are deleted 90 days after they were recorded, any copies our speech processors keep under their own terms (section 2), and anything the third-party AI model provider in sections 4 and 5 keeps under its own terms. If you had a subscription or a free trial, a record of it (plan, price, status changes, payment amounts, the cancellation reason our payment processor reports, and dates, kept to the day only) is kept after you close your account, with no link to your account, so our business figures stay accurate. It holds no name, email address or payment-processor reference. Deleting your account does not remove what is stored on your computer. The trial history record described in section 1 is created when a free trial starts, when a subscription is taken out, or shortly before a free trial ends. It is kept even after your account is deleted, because its purpose is to recognise a returning account. Each entry is deleted automatically 24 months after it was recorded.

We use a small number of carefully selected third-party services to run LotusQ. Our authentication and cloud database provider stores your account credentials and your dictionary and snippets. Our payment processor handles all billing. LotusQ never sees or stores your payment card details. Our email delivery provider sends our account emails and our service notices by email, so it receives your email address. It also sends product news to people who choose to get it. For product news, it also receives an ID that links your email address to your LotusQ account. If you send us a support request from the app, it carries that request, with your email address, to our support inbox. If you invite someone to LotusQ, it sends them your invitation, so it receives their email address along with your name, your email address and your invite link. Our website's visit counting is handled by Pirsch Analytics, provided by Emvi Software GmbH in Germany (section 1). Transcription is handled by our speech processors. We send them your audio, sometimes with your dictionary words as a vocabulary hint, only to transcribe it. Some may keep or otherwise process it afterwards under their own terms, as described in section 2. Tidying your dictated text is handled by a text processor, which receives the transcript and the name of the app you are in, never the audio. Compose requests are handled by an AI text provider, which receives the instruction, the selected text, the app name, your name for sign-offs (unless you clear it), your saved snippets (except ones you exclude) and, if you turn on On-screen reading, nearby on-screen text only to write the result. Translate requests are handled by our AI text provider, which receives only the text to translate and the language you chose, never your audio. Working out what kind of app you are dictating into is handled by a third-party AI model provider. For an app LotusQ cannot place, it receives the name of the app and, where there is one, the website host, plus the title of that window only if you turn on Use window titles to pick a better style, with email addresses, links and numbers removed first; never what you dictate or your audio (section 5). For some accounts, the same kind of provider judges auto-learn corrections, receiving only what section 4 lists and never your audio. How long the provider may keep what we send is governed by the provider's own terms; LotusQ does not ask it to keep any of it.

You have the right to access, correct, or delete any personal data we hold. You can export your dictionary as a CSV file at any time from the app; snippets cannot be exported from the app, so ask us if you want a copy. To request account deletion or a data export, contact [email protected]. We will respond within 7 days. If you ask us to erase your data, we will also delete any trial history entries we can still link to you. An entry we can no longer link to you, such as a Google sign-in entry for an account that has already been deleted, is not deleted early and expires after 24 months. If you ask us to erase your data, we also delete any subscription and trial records we can still link to you. You can also ask us to limit how we use your data. Where the law gives you the right, you can ask us for the data you gave us in a format you can take elsewhere. You have the right to object to any use of your data that relies on our legitimate interests. You can always object to the use of your data for direct marketing. If you object to product news, we stop sending it to you. Where we rely on your consent, you can withdraw it at any time; that doesn't affect anything we did before you withdrew it. If you're unhappy with how we handle your data, please tell us first at [email protected]. You also have the right to complain to the UK's Information Commissioner's Office (ICO) at ico.org.uk.

LotusQ is not directed at children under 13. We do not knowingly collect personal information from children. If you believe a child has provided us with personal data, please contact us immediately.

LotusQ is provided by ANNT CAPITAL LTD, a company registered in England and Wales (company number 16974043), with its registered office at 71-75 Shelton Street, Covent Garden, London, United Kingdom, WC2H 9JQ. We are the controller of the personal data described in this policy. For any privacy question or request, contact [email protected].

We may email the address on your account for three reasons:

• Account emails: to confirm your email address when you sign up, and to reset your password when you ask.
• Service notices: when there is an important change to this policy, our Terms of Service or your account. We may send these by email, or tell you on our website or in the app instead.
• Product news: emails about new LotusQ features, sent only if you choose to get them. It's off unless you turn it on: at sign-up, where the box starts unticked, or at any time in the app's Settings under Privacy.

You can stop product news at any time: use the unsubscribe link in any product news email, turn it off in Settings, or email [email protected]. Stopping product news doesn't stop account emails. We keep a record of your choice and when you made it.