HyperWhisper is built to keep your data on your device. You can transcribe entirely offline using local AI models, or choose from a range of cloud providers — including HyperWhisper Cloud, our built-in cloud service. Whichever option you choose, we never store your audio recordings or transcripts, with one exception for HyperWhisper Cloud audio. On the Chirp route, Cloud audio sits briefly in our own Google Cloud Storage bucket, is deleted after the request, and at most 1 day after a failure. The Chirp route is HyperWhisper Cloud's Google Chirp 3 model. A Chirp recording goes into the bucket when it is over 9.5 MB or when our server, estimating from its file size, judges it longer than about 55 seconds; other Chirp recordings are sent to Google without being written to it. Separately, when a provider returns an error to HyperWhisper Cloud, our service logs keep the provider's error message, and that message can quote part of what was sent to the provider, such as a vocabulary term or a passage of your transcript, as described in the HyperWhisper Cloud section.
Ray Amjad LTD (Company Number 14506459, incorporated in the United Kingdom) is the provider of HyperWhisper and acts as the data controller for the limited personal data we handle.
What We Collect
- Email: Used to deliver receipts, licenses, and support. We keep a record of the transactional emails we send you — the recipient address, email type, subject, send timestamp, and delivery status — retained for support, audit, and deliverability troubleshooting. We do not store the message body. When you request the download link from our website, we record the email address you submit so we can send it. We use the IP address of that request only to rate limit the form (2 requests per IP address per minute) and we do not store the IP address, browser user agent, or country with the email record. The rate limit counts requests per IP address in Cloudflare's rate limiting service; those counters expire within about a minute and hold no email address, browser user agent, or country. Our transactional emails are delivered by Cloudflare, which receives the recipient address and the message content in order to deliver them. For license key, Account Key, and sign-in link emails, that content includes the key or the sign-in link.
- Order and billing info: Processed by our payment provider (Stripe) for purchases, refunds, and fraud prevention.
- Crash and performance diagnostics: The HyperWhisper desktop apps for macOS, Windows, and Linux send diagnostics to Sentry, our error-monitoring provider. This is on by default. You can turn it off in Settings → General with the switch labelled "Error logging" on macOS and "Send error reports" on Windows and Linux. When you turn it off, the app stops collecting new diagnostics. On Windows and Linux, reports that are already waiting on your device can still be sent at that moment, and on macOS one final session record can be sent. The diagnostics include:
- Crash and error reports with technical details about the error, and other diagnostic events and records about how the app is running. On macOS, this includes routine diagnostic log records that are sent even when nothing has gone wrong.
- Release-health session data: records of app sessions and whether they ended in a crash.
- Performance traces (timings of operations in the app) on macOS and Windows, and CPU profiles (samples of which parts of the app's code were running) on macOS. The Linux app does not currently record performance traces.
- Identifiers: your IP address, the app version, your operating system version, your CPU architecture, and an identifier for your installation of the app that stays the same from one launch to the next and is linked to your sessions. On Windows and Linux, this also includes your computer's name, and on Linux the name of your user account on it.
- Context about what the app was doing. For example, this can include the names of your audio input devices and the name of the app you were dictating into with the number of characters pasted (macOS and Windows), and file paths on your computer. On macOS, a file path can contain the name of your user account.
- Server error reports: When our website or the HyperWhisper Cloud service hits an unexpected error on our servers, it sends an error report to Sentry. A report contains the type of error, a shortened error message, the places in our code where it happened, the route that failed (such as "/api/license/validate", without the address or query you requested), the request method, and, for the Cloud service, the server region. Before a report is sent, we remove Account Keys and email addresses from the error message, along with the values in a database query and structured data such as a provider's JSON response, where a transcript could appear. A report does not include your IP address, your request's headers, cookies, or body, or your audio. These reports come from our servers, not your device, so there is no setting to turn them off.
- License information: The app contacts our licensing service to activate and validate your license. During this process, we receive your Account Key, a SHA-256 hash of your device's hardware identifier (we never receive the raw identifier), and your device's hostname. This information is used to enforce our fair usage policy. No audio or transcripts are transmitted. The licensing service also sees the IP address of each activation and validation request. We use it only to rate limit those endpoints (30 requests per IP address per minute), counted in Cloudflare's rate limiting service, where the counters expire within about a minute.
- Website sign-in: When you sign in to the website, we keep an account record with your email address and the name on your account. Each sign-in also creates a session record that keeps you signed in, and it holds the IP address and browser user agent of that sign-in. A session expires 90 days after it was last refreshed, and we delete its session record within a day of it expiring.
- Custom vocabulary: When you use HyperWhisper Cloud, our servers receive your custom vocabulary terms with each transcription or post-processing request and pass them to the provider that handles it.
- HyperWhisper Cloud records: HyperWhisper Cloud keeps a cache of up to one hour, keyed by your Account Key, of whether the key is valid and its remaining credit balance, in a store hosted by Upstash. In the same store it keeps short-lived rate-limit counters keyed by your IP address and by a SHA-256 hash of your Account Key, and temporary credit holds keyed by your Account Key while a request runs. It checks the IP address of each request against a block list of IP addresses we have blocked to stop abuse, held in the same Upstash store. It writes operational logs, which can include your IP address, and Axiom stores them. Unless you turn it off, it also stores anonymous speed measurements that hold no account, no Account Key, no IP address, no audio, and no text, and deletes them after 1 year. The HyperWhisper Cloud section below describes each of these.
We do not store your audio recordings or transcripts, with one exception. On the Chirp route, Cloud audio sits briefly in our own Google Cloud Storage bucket, is deleted after the request, and at most 1 day after a failure. The Chirp route is HyperWhisper Cloud's Google Chirp 3 model. A Chirp recording goes into the bucket when it is over 9.5 MB or when our server, estimating from its file size, judges it longer than about 55 seconds; other Chirp recordings are sent to Google without being written to it. Separately, when a provider returns an error to HyperWhisper Cloud, our service logs keep the provider's error message, and that message can quote part of what was sent to the provider, such as a vocabulary term or a passage of your transcript, as described in the HyperWhisper Cloud section.
What Stays On Your Device
- Recordings: Your audio stays on your device unless you export or share it.
- Transcripts: Stored locally on your device; you control them.
- Settings and vocabulary: App preferences and optional custom vocabulary are stored locally. When you use cloud transcription, the transcription provider receives your vocabulary terms with your audio as a spelling hint. With HyperWhisper Cloud, the terms pass through our servers to that provider, and to the post-processing provider when you use HyperWhisper Cloud post-processing. When you use local models for both transcription and post-processing, no transcription or post-processing provider receives your vocabulary. On macOS, if you turn on iCloud Sync for vocabulary, Apple stores your vocabulary (words and replacements) in your iCloud account. iCloud Sync is off by default, and you turn it on separately on each Mac in the app's Vocabulary settings. When it is on, your vocabulary entries are mirrored to your own iCloud account through Apple's CloudKit service, not to a HyperWhisper account. Along with each word and its replacement, an entry carries its position in your list, when and how it was added, and an entry ID. Apple handles this data under its privacy policy. This sync covers vocabulary only: it does not send your transcripts, recordings, modes, or other settings to iCloud.
Transcription Processing
HyperWhisper supports both local (on-device) and cloud-based transcription. When using local models, no audio data ever leaves your device. The app downloads local models directly from Hugging Face (huggingface.co and the Hugging Face servers it sends downloads to), when you download a model and whenever a model's files need to be fetched again, for example after they were deleted from your device. Hugging Face receives your IP address and basic request details, such as the app's HTTP client and which model files are requested, and does not receive your audio, transcripts, vocabulary, or account information. Apple Speech on macOS is the exception: macOS downloads its speech models from Apple.
When using cloud transcription, your audio is sent directly from your device to the cloud provider you choose. HyperWhisper does not proxy, store, or retain your audio or transcripts on our servers. HyperWhisper Cloud, below, has one exception. On the Chirp route, Cloud audio sits briefly in our own Google Cloud Storage bucket, is deleted after the request, and at most 1 day after a failure. The Chirp route is HyperWhisper Cloud's Google Chirp 3 model. A Chirp recording goes into the bucket when it is over 9.5 MB or when our server, estimating from its file size, judges it longer than about 55 seconds; other Chirp recordings are sent to Google without being written to it. Separately, when a provider returns an error to HyperWhisper Cloud, our service logs keep the provider's error message, and that message can quote part of what was sent to the provider, such as a vocabulary term or a passage of your transcript, as described in the HyperWhisper Cloud section. The cloud provider you select may store or process your data according to their own privacy policies. You are responsible for reviewing those policies before use.
Third-party transcription providers (bring your own API key):
- OpenAI
- Deepgram
- ElevenLabs
- Groq
- AssemblyAI
- Mistral
- Soniox
- SpaceXAI (Grok)
- Google (Gemini). Covers both Gemini key slots, Google Gemini and Gemini 3.5 Transcribe. Both send your audio to Google.
- Meta Muse Voice Transcribe. A direct Meta request, with your own key, goes from your device to api.meta.ai and does not pass through HyperWhisper servers. A Meta request on HyperWhisper Cloud is different: it goes from our edge servers to api.meta.ai on our own Meta account, as described under HyperWhisper Cloud below.
Third-party post-processing providers (bring your own API key) — these services, and any custom endpoint you add, receive your transcribed text and your custom vocabulary terms for correction and formatting:
HyperWhisper Cloud
HyperWhisper Cloud is our built-in cloud transcription service. It routes your audio through our edge servers to one of the providers listed below. We do not store your audio or transcripts — we track credit usage and keep the operational records described below — with one exception. On the Chirp route, Cloud audio sits briefly in our own Google Cloud Storage bucket, is deleted after the request, and at most 1 day after a failure. The Chirp route is HyperWhisper Cloud's Google Chirp 3 model. A Chirp recording goes into the bucket when it is over 9.5 MB or when our server, estimating from its file size, judges it longer than about 55 seconds; other Chirp recordings are sent to Google without being written to it. Separately, when a provider returns an error to HyperWhisper Cloud, our service logs keep the provider's error message, and that message can quote part of what was sent to the provider, such as a vocabulary term or a passage of your transcript, as described in the HyperWhisper Cloud section.
Beyond credit usage, HyperWhisper Cloud keeps the operational records described in this paragraph and the next two. To check your credits without asking our licensing service on every request, it caches whether your Account Key is valid and its remaining credit balance for up to one hour, keyed by your Account Key, in a store hosted by Upstash. To limit how many requests can be sent, it counts requests in the same Upstash store, keyed by your IP address and by a SHA-256 hash of the Account Key the request sends; these counters expire within a few minutes. While a request or a live transcription session runs, it also keeps a temporary hold on the credits that request may use, keyed by your Account Key; a hold is removed when the request is settled and otherwise expires on its own. It also checks the IP address of each request against a block list held in the same Upstash store. We add an IP address to that list to stop abuse of the service.
HyperWhisper Cloud's edge servers are hosted by Fly.io. The service writes operational logs, and a log shipper we run on Fly.io sends them to Axiom, which stores them for us. A log line can include request identifiers (ours, Fly.io's, and the provider's), the server machine and region, timing, byte and character counts, the provider and model used, the language, the identifier of the mode you used, whether a specialised transcription domain such as medical was requested, your app's platform and version, credit amounts, and your IP address. We use these logs to find and fix faults, to prevent abuse, including deciding which IP addresses to block, and to keep the service running. We do not write your audio, your transcripts, your prompts, the text a model returns to you, or your Account Key into these logs. When a provider returns an error, we log the provider's error message (for some providers the whole message, for others a shortened part of it), and a provider can quote part of the request in that message, such as a vocabulary term or a passage of the text sent to it.
Unless you turn it off, HyperWhisper Cloud also stores an anonymous speed measurement in our database for each provider it calls to transcribe a recording. These measurements build our public speed comparison page. A measurement holds the provider and model, the server region, a clip-length group, the provider's place in the fallback order, how long the provider took, whether the call worked and, if it did not, the kind of failure (for example a timeout), and the hour it happened. It holds no account, no Account Key, no IP address, no audio, and no text. We delete these measurements after 1 year. Sharing is on by default. To turn it off, use the switch labelled "Share anonymous speed data" in Settings → General. Today the measurements come only from the macOS app (version 2.43.0 or later) and the Windows app (version 1.10.0 or later). Older versions have no switch, and the service does not measure them.
Transcription (speech-to-text) providers:
- Deepgram (Nova-3) — default provider. Falls back to Groq, then ElevenLabs, on failure.
- Groq (Whisper large-v3) — fastest option. Falls back to Deepgram, then ElevenLabs, on failure.
- ElevenLabs (Scribe v2) — highest accuracy option. Falls back to Deepgram, then Groq, on failure.
- Google (Chirp 3) — the Chirp route. Recordings over about 55 seconds or 9.5 MB pass through our own Google Cloud Storage bucket first, as described above.
- SpaceXAI (Grok) (Grok Voice Transcribe). Falls back to Deepgram, then Groq, then ElevenLabs, on failure.
- Microsoft (Azure Speech) (MAI-Transcribe).
- OpenAI (its GPT transcription models).
- Google (Gemini) (Gemini Flash and Pro models).
- Google (Gemini 3.5 Transcribe).
- AssemblyAI (Universal-3.5 Pro, Universal-2 and Dictation).
- Mistral (Voxtral).
- Soniox (Soniox async speech-to-text).
- Meta (Muse Voice Transcribe). Our edge servers send your audio to api.meta.ai on our own Meta account.
When HyperWhisper Cloud transcribes a recording with Deepgram, Groq, ElevenLabs or SpaceXAI (Grok), the same audio, and any custom vocabulary sent with it, can go to the next provider in that provider's fallback chain. This happens when the provider handling the request times out, returns a server error or a rate limit, returns a reply we cannot use, rejects the audio or a setting such as the language, refuses our account key, or cannot be reached from the server region that received your request. It also happens once when the provider you chose returns an empty transcript for audio it says it processed. The chains are: Deepgram, then Groq, then ElevenLabs; Groq, then Deepgram, then ElevenLabs; ElevenLabs, then Deepgram, then Groq; and SpaceXAI (Grok), when you choose it, then Deepgram, then Groq, then ElevenLabs. So one request can reach at most these four providers. Every other HyperWhisper Cloud transcription provider, Google (Chirp 3) included, has no fallback and never passes your audio to another provider. You are billed only for the provider that returned your transcript.
Post-processing (text correction) providers:
- Cerebras (GPT-OSS-120B) — default provider. Falls back to Groq on failure.
- Groq (GPT-OSS-120B) — fallback provider. Falls back to Cerebras on failure.
- Anthropic (Claude) (Claude Haiku 5.5). Falls back to Cerebras on failure.
- SpaceXAI (Grok) (Grok 4.3). Falls back to Anthropic on failure.
- OpenAI (GPT-5 mini or GPT-5 nano). Falls back to Anthropic on failure.
- Google (Gemini) (Gemini 2.5 Flash, Gemini 2.5 Flash-Lite or Gemini 3.8 Flash). Falls back to Cerebras on failure.
- Mistral (Mistral Small). Falls back to Groq on failure.
These providers receive your transcribed text, and the context described under "Screen Text and app context" below, to correct and format it. This post-processing request carries no audio. Grok and Mistral store request data for approximately 30 days for abuse monitoring, and do not use it to train their models. A request goes to the fallback named above in only four cases: the request to the provider still fails with a server error after any automatic retries (this includes a reply we cannot read), the provider does not answer within our time limit, the provider rejects the request because of its rate limit (in these last two cases the request goes to the fallback at once, with no retry on the first provider), or its reply repeats our instructions instead of your corrected text. So one request can reach at most three providers in this list — for example Grok, then Anthropic after a server error, then Cerebras if Anthropic's reply repeats the instructions.
Deepgram is the one HyperWhisper Cloud provider account on which we have configured zero data retention: Deepgram's data retention is set to zero so that audio is deleted immediately after transcription. The other HyperWhisper Cloud providers listed above keep data under their standard API terms, and we do not claim a zero-retention or no-training setting on their accounts. We are not on enterprise plans with these providers, so their standard data processing terms still apply. Please review each provider's privacy policy linked above for full details on how they handle data.
Under those standard terms, ElevenLabs stores your audio and transcript by default, because its zero-retention mode is available only on enterprise plans and our account is not on one yet. Grok and Mistral store transcription requests for approximately 30 days for abuse monitoring, and do not use them to train their models. Meta does not publish a retention statement for its Muse speech-to-text API, so we do not claim one for Meta. Our Data Privacy help page lists the control we apply for each provider.
HyperWhisper is fully open source under Apache-2.0 — including the Cloud backend. You can read the full source code for the Cloud backend and the apps on GitHub.
Medical Mode
With HyperWhisper Cloud, a mode that uses AssemblyAI can be set to the medical Transcription Domain. Our edge servers then send the audio to AssemblyAI with its Medical Mode add-on turned on. AssemblyAI processes it on HyperWhisper Cloud's own AssemblyAI account, under AssemblyAI's privacy policy, and we do not promise zero data retention there. A HyperWhisper Cloud mode can also use Deepgram's Nova-3 Medical or Nova-2 Medical model. These options tune transcription for clinical vocabulary. For these HyperWhisper Cloud medical requests, our edge servers do not write the audio or the transcript to our own storage, although a provider's error message in our logs can quote part of the request, as described in the HyperWhisper Cloud section. If Deepgram cannot take a request, HyperWhisper Cloud can send it to Groq or ElevenLabs as described above, and they do not use a medical model. Live streaming through HyperWhisper Cloud does not use a medical model. With your own AssemblyAI or Deepgram API key, you can instead choose AssemblyAI's Universal-2 Medical or Universal-3.5 Pro Medical model, or Deepgram's Nova-3 Medical or Nova-2 Medical model (Nova-3 Medical also for live streaming); your audio then goes from your device directly to that provider, under your own account, and not through our servers. With your own ElevenLabs API key, you can also choose ElevenLabs' Scribe v2 Medical model; your audio and your custom vocabulary then go from your device directly to ElevenLabs, under your own account, and not through our servers. If post-processing is on, the transcript also goes to the post-processing provider you use. Each provider processes your data under its own terms, and you must review them before you dictate about a patient.
Medical Mode is a vocabulary and formatting feature only. We are not a healthcare provider, and we do not sign a Business Associate Agreement (BAA). HyperWhisper, including HyperWhisper Cloud and Medical Mode, is not offered as HIPAA-compliant or as fit for regulated clinical record keeping.
Screen Text and app context
The post-processing request sent to the provider you choose (HyperWhisper Cloud, a bring-your-own-key provider, or a custom endpoint you add) carries more than your transcribed text. It includes your computer's name, local time, time zone and locale. It can also include the name of the app in front, the window or browser tab title, the website host, the type and label of the focused field, and up to 100 characters of the text in that field or of the text you have selected. A mode can also turn on Screen Text, which is off by default in every mode. When a recording starts, the app reads the text on the display that shows your active window (on macOS, the main display when it cannot find that window; on Linux, the area you select in the screenshot prompt) and adds up to 2,000 characters of it to the request. Text recognition runs on your device: the screenshot itself is never sent, only the recognized text. On macOS and Windows, no screenshot is taken when HyperWhisper itself is the app in front. When HyperWhisper identifies the app in front as sensitive, such as a password manager, the screen text and the field or selected text are left out, but the app name, the window or tab title, the website host and the field type and label are still sent. When post-processing is off, none of this is sent to a post-processing provider.
Payments and Licensing
- Payments: All payments are handled by Stripe. We receive the minimum order metadata required to fulfill your purchase and provide support. Please review the Stripe Privacy Policy for details on how they handle payment data.
- Past payments (before December 2025): Payments made before December 2025 were processed by Polar and their processors (e.g., Stripe). If you purchased through Polar, your order data may still be retained by them. See the Polar Privacy Policy for details.
- Licensing: License activation/validation may send your Account Key and minimal device information to our licensing service to prevent abuse. No audio or transcripts are transmitted.
Cookies and Analytics
The app does not set tracking cookies. Our website uses strictly necessary cookies for authentication (when you log in to your account to manage your subscription or add credits), checkout, and license management. We do not use advertising cookies.
We use PostHog for analytics to understand how our website is used and to improve the experience. PostHog may set its own cookies. We do not use this data for advertising purposes. Our website also sends PostHog an error report when a page hits an unexpected error, for example while you sign in, buy credits, or load a download page, so we can find and fix it. A report holds the error type and message, the stack trace, the page address, and your browser and device details, and PostHog links it to the same visitor record as our analytics. When our own code catches the error, the report also names the step that failed and can add details such as the download platform you picked or a failed request's status code. We do not add your email address, your Account Key or a sign-in link to a report, and an Account Key that appears in a report is removed before it is sent.
The demo video on our home page is hosted by Bunny.net (Bunny Stream) and plays in an embedded player from iframe.mediadelivery.net. When the video player loads, Bunny.net receives your IP address, your browser user agent, and our website's address. The player may record plays and watch time, and it may store your playback settings, such as volume, in your browser's local storage. We do not send Bunny.net your email address, your account, your license (Account Key), your audio, or your transcripts through the player.
The desktop apps use Sentry, our error-monitoring provider, for crash and performance diagnostics, as described under What We Collect. You can turn it off in Settings → General. Our website and the HyperWhisper Cloud service also send server error reports to Sentry, as described under What We Collect.
Our website pages embed an AI support chat widget served by AgentStack (agentstack.build), another product by Ray Amjad. When the widget loads, that service receives your IP address, your browser user agent, and our website's address. Anything you type into the chat is sent to that service. The widget also loads its own analytics, a separate PostHog instance on AgentStack's domain (t.agentstack.build), which may set its own cookies or local storage. The widget has no access to your audio, your transcripts, or the desktop app.
Our documentation at hyperwhisper.com/docs is hosted by Mintlify. When you read a docs page, Mintlify records the page view: the page address and title, the referrer, your browser user agent, your screen and window size, your browser language, your time zone, and your network type. It keeps a random identifier in your browser's local storage so that it can link your visits, and it sends a session identifier with each page view. Mintlify receives your IP address. When you use the docs search, what you type into the search box is sent to Mintlify to find matching pages. The search uses hCaptcha, a bot-detection service: when you open the search, your browser loads hCaptcha, which receives your IP address and details about your browser. The docs pages have no access to your audio, your transcripts, or the desktop app.
Children's Privacy
HyperWhisper is not directed to children under 16. We do not knowingly collect personal information from children. If you believe a child has provided us information, contact us to delete it.
Your Rights
Because we do not store your audio or transcripts, requests to access or delete that data should be carried out on your device by you. The one exception is HyperWhisper Cloud audio. On the Chirp route, Cloud audio sits briefly in our own Google Cloud Storage bucket, is deleted after the request, and at most 1 day after a failure. The Chirp route is HyperWhisper Cloud's Google Chirp 3 model. A Chirp recording goes into the bucket when it is over 9.5 MB or when our server, estimating from its file size, judges it longer than about 55 seconds; other Chirp recordings are sent to Google without being written to it. Separately, when a provider returns an error to HyperWhisper Cloud, our service logs keep the provider's error message, and that message can quote part of what was sent to the provider, such as a vocabulary term or a passage of your transcript, as described in the HyperWhisper Cloud section. For email, order, or license information we maintain, you may request access or deletion by contacting support. Some information must be retained for legal/accounting purposes.
If you dictated patient information with the Medical Mode options described above, through HyperWhisper Cloud or your own API key, we hold neither the audio nor the transcript, apart from any part of a HyperWhisper Cloud request that a provider's error message quotes in our logs, so delete them on your device as described above. Any copy a provider keeps is governed by that provider's terms.
Hosting
Our website, the account area, and our licensing service are hosted by Cloudflare. Cloudflare runs the code that handles each request to them, so it receives the request's IP address and browser or app details, and what the request carries: for example your Account Key and device details in a license activation or validation request, the email address you submit in a form, or the sign-in link you open.
HyperWhisper Cloud runs on servers hosted by Fly.io. Fly.io runs the code that handles each HyperWhisper Cloud request, so it receives the request's IP address and everything the request carries: for example your audio or text, your mode's instructions, your Account Key, your custom vocabulary, the language and mode you chose, and your app's platform and version. Its operational logs are stored by Axiom, as described under HyperWhisper Cloud above.
Our database is hosted by PlanetScale. It holds these account and license records: your email address and the name on your account, your Account Key, your order and customer ids, your HyperWhisper Cloud credit balance and the history of the credits you bought or were granted (amounts, refunds, expiry dates, and the payment or grant they came from), the name of each device you activate (on macOS the computer name, which often includes your first name; on Windows and Linux the computer's network name), its hashed device identifier, and the dates it was first and last validated, the email addresses submitted for the download link, the records of the transactional emails we send you, your sign-in session records, which include the IP address and browser user agent of the sign-in, and a record of each sign-in link you request (your email address and the link's token), which can remain after the link expires.
International Transfers
Our payment and licensing providers, Upstash, Cloudflare, and PlanetScale may process limited personal information in multiple countries, including the United Kingdom, the European Economic Area, and the United States. Upstash hosts HyperWhisper Cloud's rate-limit store. It also holds a one-hour cache, keyed by your Account Key, of whether the key is valid and its remaining credit balance. HyperWhisper Cloud checks request IP addresses against a block list held there. See the Upstash Privacy Policy for details. Where required, appropriate safeguards are used by those providers, including Upstash (for example, adequacy decisions, Standard Contractual Clauses, or equivalent mechanisms). The AgentStack support chat widget described above is served from the United States. Mintlify, which hosts our documentation as described above, may process the docs page views it records and the docs search queries it receives in the United States. Hugging Face, which serves local model downloads as described under Transcription Processing, may process those requests in the United States and other countries. Fly.io runs HyperWhisper Cloud's edge servers in regions around the world, so a Cloud request can be handled outside the United Kingdom and the European Economic Area, and Axiom may process the service's logs in the United States.
Changes to This Policy
We may update this policy. If we make material changes, we will update the date above and, where appropriate, notify recent purchasers.
Contact
Data Controller: Ray Amjad LTD (Company Number 14506459, United Kingdom). Questions or requests: hi@support.hyperwhisper.com
