Privacy Policy

Mandra Voice Version 1.1 · Effective 18 September 2026 · Last updated 27 September 2026

YAANTRIX Technologies Inc. ("Mandra Voice", "we", "us", "our") operates the Mandra Voice desktop application and the website at mandravoice.com (together, the "Service"). This policy explains what personal information we collect, why, who we share it with, how long we keep it, and the rights you have over it.

We are the controller of your personal information (under PIPEDA, the "organization" accountable for it). We are incorporated in Ontario, Canada, with a registered office at 4600 Joy Rd, Maidstone, Ontario N0R 1K0, Canada.

• Privacy Officer: Hardik Gora — support@mandravoice.com
• Grievance contact for India (DPDP Act, 2023): Hardik Gora — support@mandravoice.com

Summary

The things people most want to know, stated plainly. Everything here is expanded below, and each statement describes how the software actually behaves.

• Your microphone is off until you press your hotkey. Mandra Voice captures audio only while you hold or toggle your dictation hotkey, or click the record button in the widget. There is no background listening, no wake word that listens continuously, and no always-on recording mode.
• We do not keep your audio. Audio is streamed to a speech-to-text provider, converted to text, and discarded. We never write audio to disk, never place it in object storage, and never keep a copy after your dictation ends.
• We do not create voiceprints. We do not generate, store, derive, or use voiceprints, speaker embeddings, speaker-identification models, or any other biometric identifier from your voice, and we do not permit our providers to do so.
• We do not sell or share your personal information, as "sell" and "share" are defined under the California Consumer Privacy Act, and we never have.
• We do not train AI models on your data, and our contracts prohibit our providers from training on it either.
• You control your transcripts. They are kept for 30 days by default. You can change that to 7 days, 90 days, 1 year, or forever — or turn on Privacy Mode so no transcript is ever written down at all.
• You can take your data or delete it at any time, from Settings → Data & Privacy, without emailing anyone.

1. Personal information we collect

1.1 Information you give us

Account information. Your email address, and — if you provide them — your first name, last name, and profile picture. If you sign in with Google we receive your email address, name, and profile picture from Google; we never receive your Google password. We store a country code derived from your account settings so we can apply the right tax and consumer rules.

Authentication data. A password hash (we never see or store your password itself), or one-time email codes, or a Google OAuth identifier — depending on how you sign in. Authentication is handled by Supabase Auth on our behalf.

Agreement records. When you accept these terms we record the version accepted, the timestamp, your IP address, and your browser or app user agent. This is the evidence that you agreed, and we are required to be able to produce it. Lawful basis: legitimate interests (GDPR Art. 6(1)(f)) in being able to prove contract formation.

Marketing consent records. If you opt in to product and promotional email — at signup or later in Settings → Notifications — we record that decision, when you made it, where you made it, the exact wording you were shown, and your app or browser user agent. We record withdrawals the same way. Canada's Anti-Spam Legislation puts the burden on us to prove you consented, and this is how we do it. We keep these records for three years after your last decision, as CASL requires.

Support correspondence. Anything you send us by email, including the content of your message and any attachments.

1.2 Information created when you dictate

Voice audio. Captured only while dictation is active. Streamed in real time (or, for the batch transcription engine, uploaded as a short clip) to the speech-to-text provider configured for your account. Not retained by us at any point — not on your device, not on our servers, not in backups. Under the CPRA, voice recordings are sensitive personal information; see §6.4 for the corresponding right.

Transcripts. The raw text returned by the transcription engine and the processed text that gets inserted into your application, together with the language detected, the dictation duration, the word count, the context mode, and the timestamp. Stored in your account unless Privacy Mode is on. Subject to the retention period you choose (§5).

Insertion activity. What text was inserted, a 100-character preview of it, whether you inserted, copied, dismissed, corrected, or undid it, and the name of the application it went into (for example Slack, Mail, Visual Studio Code). We use the application identity to choose an appropriate writing style — an email is formatted differently from a code comment — and to power your history view. We do not read the contents of your applications, and we do not see anything you did not dictate.

Your customisations. Custom dictionary words, snippets, per-application overrides, selected languages, and your hotkey configuration.

1.3 Information we collect automatically

Usage metering. Word counts, dictation counts, recording seconds, and Read Aloud counts, aggregated per day and per week. We need these to enforce the free-tier limit and to bill accurately. Word counts continue to be recorded in Privacy Mode — a count is not content.

Device and session information. A device identifier we generate, a device name, operating system platform, and app version, plus the time your device was last seen. This enforces the device limits in §11 of the Terms (3 registered devices, 2 running at once) and lets you see and revoke sessions.

Service operation logs. Which transcription or AI provider handled a request, how long it took, whether it succeeded, an error message if it failed, the audio duration, and the computed cost. These contain no transcript content.

Abuse prevention. Counts of requests per endpoint per time window, so we can rate-limit credential-stuffing and quota abuse.

Diagnostics and crash reports. When the app crashes or hits an unexpected error we send a report to Sentry containing the error, a stack trace, your app version, your operating system, and the sequence of actions leading up to it. We scrub these reports before they are sent, and they are not designed to carry transcript text. If your browser or system sends a Global Privacy Control signal, we automatically disable performance tracing and keep only crash capture (§8).

Email delivery records. Which email type we sent you and when, plus the delivery provider's message identifier — so we do not send you the same reminder twice.

1.4 Billing information

We never see your card. Card numbers, CVCs, and expiry dates go directly from your browser to Stripe and never touch our servers. From Stripe we receive and store your Stripe customer identifier, subscription identifier, plan, status, current billing period, and — for display in the app only — your card brand and last four digits. Stripe's own handling of your payment data is governed by Stripe's Privacy Policy.

1.5 Information we do not collect

To be explicit, because voice products are often assumed to do these things:

• We do not record or transcribe audio when dictation is not active.
• We do not capture your screen, read your clipboard except to paste text you dictated, or log your keystrokes. The keyboard hook we install watches for one thing: whether your configured hotkey is pressed or released.
• We do not read the contents of the applications you dictate into. We detect only which application is in front.
• We do not generate or store voiceprints, speaker embeddings, or biometric identifiers.
• We do not use advertising cookies, tracking pixels, or third-party analytics scripts.
• We do not buy personal information about you from data brokers.
• We do not make automated decisions that produce legal or similarly significant effects about you (GDPR Art. 22). Nothing about your account is decided by an algorithm without a person involved.

2. Why we process it, and our lawful basis

For users in the EEA, the UK, and Switzerland, GDPR requires us to name a lawful basis for each purpose. Users elsewhere may find the table a useful plain statement of what we do and why.

Where we rely on legitimate interests we have carried out a balancing test and concluded that our interest does not override your rights, largely because the data involved is narrow, is not used to profile you, and is not shared for anyone else's purposes. You can ask us for the assessment at support@mandravoice.com, and you can object under §6.

Consent under Canadian law. PIPEDA requires meaningful consent for collection, use, and disclosure. Installing Mandra Voice, granting microphone permission, and pressing your hotkey to dictate constitute express consent for the transcription purposes described here. You can withdraw that consent at any time by turning off the relevant feature, revoking microphone permission in your operating system, or closing your account — though doing so means the Service can no longer function.

3. How your voice is processed

This section describes exactly what happens to the sound of your voice, because it is the most sensitive thing the Service touches.

1. You press and hold (or toggle) your dictation hotkey. Only then does the microphone stream open.
2. Audio is encoded and sent over an encrypted connection (TLS/WSS) to the speech-to-text provider selected for your account. Depending on configuration this is Groq (Whisper transcription, the default, with OpenAI as its backup), Deepgram, or ElevenLabs. If your chosen provider fails mid-session, the app fails over automatically to another so your dictation is not lost — the alternate provider is disclosed in §4.
3. The provider returns text. The audio is discarded.
4. The text is cleaned locally on your device (filler words, punctuation, capitalisation).
5. If AI enhancement is on, the text — not the audio — is sent to our text-improvement provider (§4) and a polished version comes back.
6. The final text is typed or pasted into whichever application is in front.
7. A transcript row is written to your account, unless Privacy Mode is on.

Where processing happens. Audio is processed by our transcription provider in the United States, and the AI text features likewise. (The app still routes Deepgram traffic to that provider's European endpoint when your device is set to a European time zone, a leftover from when we served those countries; we do not offer the Service in the EEA or the UK — see §6.3.)

Retention by providers. We use each transcription and AI provider through its API under that provider's published terms, and none of them may use your audio or text to train their models. That is not the same as zero retention: a provider may hold content briefly for abuse monitoring or operational reasons under its own terms — OpenAI, for example, retains API inputs and outputs for up to 30 days unless zero-retention is enabled on the account. Where a provider offers logging as a default we disable it: for ElevenLabs we set the enable_logging=false flag on every request, which opts out of their default retention and training pathway. Except where we say otherwise, we rely on these published API terms rather than an individually negotiated zero-retention agreement. If none of your dictation should reach our servers at all, use Privacy Mode (below); the audio still goes to the transcription provider, because that is how the text is produced.

Privacy Mode. In Settings → Data & Privacy you can enable Privacy Mode. With it on, no transcript is written to your account at all — dictation still works, text is still inserted, and word counts are still recorded for billing, but the text itself exists only transiently in memory and in the app you dictated into. This is the strongest privacy setting the Service offers, and it is available on every plan including the free one.

4. Who we share it with

We disclose personal information only to the service providers listed on our Subprocessors page, each of which is contractually bound to process it only on our instructions, to keep it confidential and secure, and not to use it for their own purposes or to train models.

The current list, in summary:

We will notify you at least 30 days before adding a new subprocessor. If you object, you may cancel before the change takes effect and we will refund the unused portion of your term.

Beyond those providers we disclose personal information only:

• To you, or to anyone you direct us to.
• To your organisation's administrator, if you use Mandra Voice under an Enterprise plan that your employer pays for. They can see your usage and manage your access; they cannot read your transcripts.
• When the law requires it — to comply with a valid subpoena, court order, warrant, or other lawful demand. We will notify you before disclosing unless we are legally prohibited from doing so, and we will resist requests that appear overbroad or unlawful.
• To protect people — where disclosure is necessary to investigate fraud, prevent imminent harm, or enforce our Terms and Acceptable Use Policy.
• In a business transfer — if we are acquired, merged, or reorganised, personal information may transfer to the successor, who will remain bound by this policy or a materially equivalent one. We will notify you before your information becomes subject to a different policy.

We do not sell or share personal information. We have never sold personal information, and we do not share it for cross-context behavioural advertising, as those terms are defined by the CCPA and CPRA. We do not disclose personal information to third parties for their own marketing.

5. How long we keep it

6. Your rights

6.1 Everyone, wherever you live

Regardless of the law that applies to you, we give every user the same core controls:

• See your data — Settings → Data & Privacy → Export data produces a PDF containing your profile, transcripts, and snippets. No request, no waiting.
• Correct your data — edit your profile, snippets, and dictionary in the app.
• Delete your data — Settings → Data & Privacy → Delete account permanently erases your profile, transcripts, activity log, snippets, dictionary, notifications, referrals, usage rows, and device sessions, cancels any active subscription, and removes your login. This cannot be undone. Billing records are retained as described in §5.
• Delete some of it — reset your statistics, shorten your retention window (which immediately deletes anything older), or turn on Privacy Mode.
• Opt out of optional processing — turn off usage analytics, AI enhancement, or reminder emails at any time.

To exercise any right that is not self-service, email support@mandravoice.com. We verify identity by requiring the request to come from the account's registered email address; for higher-risk requests we may ask you to confirm from within the app. We respond within 30 days and will tell you if we need more time.

We will never charge you, deny you service, give you a worse price, or reduce the quality of the Service because you exercised a privacy right.

6.2 Canada (PIPEDA, and provincial laws in Quebec, Alberta and British Columbia)

You may ask us for access to your personal information, ask us to correct it, withdraw consent (subject to legal and contractual restrictions), and be told to whom it has been disclosed. We will tell you if we cannot fulfil a request and why.

If you are not satisfied with our response, you may complain to the Office of the Privacy Commissioner of Canada (priv.gc.ca, 1-800-282-1376), or to your provincial regulator — the Commission d'accès à l'information du Québec, the Office of the Information and Privacy Commissioner of Alberta, or the Office of the Information and Privacy Commissioner for British Columbia.

Quebec residents (Law 25). In addition: you have the right to data portability in a structured, commonly used technological format (the export tool satisfies this); the right to be informed when personal information is used to render a decision based exclusively on automated processing (we do not do this); and the right to have information de-indexed in certain circumstances. Our privacy officer is Hardik Gora, reachable at support@mandravoice.com. We maintain a confidentiality-incident register and assess privacy risks before transferring personal information outside Quebec.

6.3 European Economic Area, United Kingdom, Switzerland (GDPR / UK GDPR)

We do not offer the Service in the EEA or the UK (see our Terms, §2). This section still applies to anyone in those regions whose personal information we hold — for example, someone who signed up before that restriction, or who contacted us.

You have the rights of access (Art. 15), rectification (Art. 16), erasure (Art. 17), restriction (Art. 18), portability (Art. 20), and objection (Art. 21) — including an absolute right to object to direct marketing. Where processing is based on consent, you may withdraw it at any time without affecting the lawfulness of what came before.

You have the right to lodge a complaint with your national supervisory authority. A list is at edpb.europa.eu; in the UK it is the ICO (0303 123 1113).

International transfers. Your account data, transcripts, and settings are stored in Canada (our database runs in the AWS Canada Central region). Speech, AI, payment, email, and error-reporting providers process data in the United States. The European Commission has determined that Canada provides an adequate level of protection for personal data transferred to commercial organisations subject to PIPEDA (Decision 2002/2/EC), which covers transfers to us. Onward transfers to our providers in the United States rely on the EU–US Data Privacy Framework where the provider is certified, and otherwise on the European Commission's Standard Contractual Clauses together with supplementary technical measures — encryption in transit and at rest, and contractual prohibitions on retention and model training. The transfer mechanism for each provider is named on our Subprocessors page. You may request a copy of the clauses at support@mandravoice.com.

Providing your personal information is a contractual necessity: without an email address we cannot create an account, and without microphone access the Service cannot transcribe. You are not obliged to provide a name, a profile picture, or optional analytics.

6.4 California (CCPA / CPRA) {#do-not-sell-or-share}

In the twelve months before this policy's effective date we collected the following statutory categories of personal information, from you and from your device, for the business purposes set out in §2, and disclosed them to the service providers listed in §4:

We have not sold or shared personal information, and we do not sell or share it now. We have not sold or shared the personal information of consumers under 16, because we do not knowingly permit anyone under 13 to use the Service at all.

You have the right to know, delete, correct, opt out of sale or sharing (there is nothing to opt out of — we do neither), limit the use of sensitive personal information, and not be discriminated against for exercising these rights.

Limiting sensitive personal information. We use your voice recordings solely to perform the transcription you requested, and for no secondary purpose — not to infer characteristics about you, not for advertising, not for research. That use falls within the exemptions in Cal. Civ. Code §1798.121(a), so no separate limitation right is triggered. If you want to restrict it further anyway, turn on Privacy Mode or stop dictating.

Global Privacy Control. We automatically honour the GPC signal. See §8.

Authorised agents may submit requests on your behalf with written permission signed by you; we will ask the agent for that permission and may ask you to confirm directly.

We respond to verifiable requests within 45 days, extendable once by a further 45 days with notice.

6.5 Other US states

Residents of Virginia, Colorado, Connecticut, Utah, Texas, Oregon, Montana, and other states with comprehensive privacy laws have rights of access, correction, deletion, portability, and opt-out of targeted advertising, sale, and profiling. We do not engage in targeted advertising, sale, or profiling for significant decisions. Use the same contact and self-service tools above. Where your state provides an appeal process, you may appeal a denied request by replying to our response; if we deny the appeal we will tell you how to complain to your state's Attorney General.

6.6 Illinois — biometric information

For the avoidance of doubt under the Biometric Information Privacy Act: Mandra Voice does not collect, capture, purchase, receive, or otherwise obtain a "biometric identifier" or "biometric information" as those terms are defined in 740 ILCS 14/10. We do not create voiceprints. We do not derive speaker embeddings. We do not perform speaker identification or verification. Audio is converted to text and discarded, and the text is not a biometric identifier.

6.7 Elsewhere

Users in Australia, New Zealand, India, Brazil, Japan, South Korea, South Africa, and other jurisdictions with data-protection statutes may exercise equivalent rights by contacting support@mandravoice.com. We apply the standards in this policy globally and do not offer a weaker standard to anyone based on where they live.

7. Security

• In transit: every connection to our servers and to our providers uses TLS 1.2 or higher; audio streams use secure WebSockets.
• At rest: databases and backups are encrypted.
• Access control: every table enforces row-level security in the database itself, so a bug in the application cannot expose one user's rows to another. Administrative access is restricted, audited, and requires separate authorisation.
• On your device: if you supply your own provider API keys, they are encrypted using your operating system's secure storage — the macOS Keychain, or Windows DPAPI — not written to a plain-text file. The desktop app runs with context isolation on and Node.js integration off, so a compromised web view cannot reach your file system.
• Secrets: provider API keys live in server-side secret storage and are never included in the application you download. Requests to providers are proxied through our backend precisely so that the keys never reach your device.
• Payments: card data goes directly to Stripe (PCI-DSS Level 1) and never touches our systems.

No system is perfectly secure. If we become aware of a breach affecting your personal information we will notify you and the relevant regulators as required — under PIPEDA where there is a real risk of significant harm, under GDPR Art. 33/34 within 72 hours, and under Quebec Law 25 for confidentiality incidents.

We do not currently hold a SOC 2, ISO 27001, or HIPAA attestation. Do not use Mandra Voice to dictate protected health information; see the Acceptable Use Policy.

8. Cookies, tracking, and Global Privacy Control

The desktop app is not a website and sets no cookies. Our website and admin portal use only strictly necessary cookies — the ones that keep you signed in. We use no advertising cookies, no tracking pixels, no session replay, and no third-party analytics.

We honour the Global Privacy Control signal automatically. If your browser or extension sends Sec-GPC: 1, we disable performance tracing in our error reporting (we still capture crashes, so bugs affecting you still get fixed) and we switch your usage-analytics preference off. You do not have to do anything, and there is no cookie banner because there is nothing to consent to. Learn more at globalprivacycontrol.org.

Full detail is in our Cookie Policy.

9. Email

Service emails — receipts, payment failures, password changes, plan changes, security notices, and terms updates — are part of your account and cannot be turned off while you have one.

Re-engagement and product emails — the "we haven't seen you in a while" messages and usage summaries — are commercial electronic messages under Canada's Anti-Spam Legislation. We send them only with your consent, every message identifies us and carries a working unsubscribe link that we honour within 10 business days (usually immediately), and you can also switch categories off in Settings → Notifications.

10. Children

Mandra Voice is not for children. You must be at least 13 to use it, and at least 16 if you are in the EEA or the UK, unless your country sets a lower digital-consent age at which point that age applies. We ask you to confirm this at signup.

We do not knowingly collect personal information from anyone below those ages. If you believe a child has given us personal information, contact support@mandravoice.com and we will delete the account and its data promptly.

Do not use Mandra Voice to transcribe the voice of a child without the consent of their parent or guardian.

11. Changes to this policy

We may update this policy as the product and the law change. When a change is material we will update the version and date at the top, email your registered address at least 14 days before it takes effect, and show a notice in the app. Continuing to use the Service after the effective date means you accept the updated policy; if you do not, close your account before that date. Previous versions are archived and available on request.

12. Contact us

If you are unsatisfied with our response you may complain to the Office of the Privacy Commissioner of Canada, your provincial privacy commissioner, your EEA or UK supervisory authority, or your state Attorney General, as applicable.