ZapVox

Transcribe, translate, and summarize WhatsApp Web audio messages with AI

Supplier identification

TECHMED SERVIÇOS MÉDICOS LTDA — CNPJ 46.367.941/0001-69
Rua Prof. Rômulo Almeida, 52, Ed. Lafayette Velloso — Acupe de Brotas — Salvador/BA — CEP 40290-030 — Brasil
Contact: [email protected] · zapvoxia.com.br

ZapVox is operated by the legal entity above, responsible for the contractual relationship, billing and support, and the controller of the personal data processed in the Service (Brazilian LGPD, art. 5, VI).

1. Introduction

This Privacy Policy describes how the ZapVox extension ("we", "us", "our") collects, uses, and protects user information ("you"). ZapVox is a browser extension for Chrome/Edge that transcribes, translates, and summarizes audio messages on WhatsApp Web.

Last updated: September 5, 2026

2. Data We Collect

2.1 How each type of data is handled

Audio messages: Audio is sent for transcription to our cloud service — which tries the providers in this order: Groq → Google (Gemini) → Cloudflare (Workers AI) → OpenAI, moving to the next one only when the previous one fails — or directly to Groq/OpenAI when you use your own API key (BYOK); in BYOK mode the audio does not pass through our servers. Raw audio is never stored to disk. The full list is in the "AI providers (transcription)" section.
Reminders (local): Reminders you create are stored in your browser and fire locally via chrome.alarms. On the Pro plan you can optionally sync them to your account so they also fire on your other devices (see section 2.2).
Cached transcriptions: Transcribed text is stored in chrome.storage.local for up to 90 days or until the cache reaches 7 MB — whichever comes first. When either limit is hit, the oldest transcriptions are deleted automatically to make room. This avoids reprocessing and powers the search across your transcriptions. This data stays exclusively in your browser.
Settings: Preferences such as language, feature toggles, and hardware profile are stored locally. One single exception: if you configure your own key (BYOK), the server stores a yes/no saying it exists — never the key, never the provider, never the usage (see the table in section 2).
Extension preferences: Your chosen transcription mode, AI model, and other preferences are stored exclusively in chrome.storage.local and never sent to servers.
Anonymous usage statistics: Local counters (e.g., total transcriptions, daily usage) are stored in your browser to display in the extension panel. This data is never transmitted externally.
Deleted messages — DISCONTINUED feature: Recovery of deleted messages has been disabled and is no longer available in any version of ZapVox. The extension does not store deleted messages, and the local database that used to exist (zapvox_deleted_msgs) is no longer created or written to. If you used an older version, that database was purged automatically under the 7-day window that already applied.
Local cascade telemetry (opt-in, v3.4.0+): If enabled by the user, diagnostic counters for the transcription cascade (how many times each cascade Phase resolved the audio) are stored locally in chrome.storage.local. This data never leaves your browser — it is used only for you to review on the popup's Diagnostics screen. No message identifier, contact name, content, or audio metadata is recorded. The telemetry is strictly a whitelist of permitted technical phases. The "Local diagnostic telemetry" toggle controls collection and the "Clear" button resets the counter.
Sentiment analysis cache (v3.4.9+): When you click the 🎭 Sentiment button, the analysis result is stored in chrome.storage.local (key zapvox_sentiment_cache) under the same 90-day and 7 MB limits, so the sentiment chip survives page reloads without reprocessing. The associated message identifier is the same sanitized hash used by the transcription cache. This data never leaves your browser and is automatically purged by the same 90-day and 7 MB limits.

2.2 Data sent to servers (only when necessary)

Data	When	Destination	Purpose
Audio (base64)	Cloud transcription	Supabase Edge Function → Groq → Google (Gemini) → Cloudflare (Workers AI) → OpenAI, in this order, moving to the next one only when the previous one fails	Convert audio to text
Audio (base64)	BYOK transcription (your own key)	Groq or OpenAI (directly from browser, without passing through us)	Convert audio to text
Conversation names (the name saved in your address book and the contact's profile name, as shown in your WhatsApp Web)	Along with the audio, on every transcription — if the "Conversation names in the transcription" switch is on (it ships on)	The same destination as the audio: our Edge Function → transcription provider, or Groq/OpenAI directly under BYOK	Spelling hint for the model (Whisper's context field), so the speaker's name comes out spelled as it is in your WhatsApp. They are read at the moment of each transcription, straight from your WhatsApp Web (which is why renaming a contact already counts for the next audio), sent along with the audio and only at that moment, and not recorded on our servers or anywhere else because of this feature — the provider processes them with the audio and discards them, under the same retention rules as the audio. The only record of names that exists is the 30-day local cache described in 2.3, which already existed and does not change.
Transcribed text	Translation or summary	Supabase Edge Function → Groq/Gemini	Translate or summarize text
JWT token	Authentication	Supabase Auth	Validate session and plan
Email	Login/signup	Supabase Auth	Create and manage account
Gender and profession (optional, with "prefer not to say")	Only if you fill them in under My account → About you	Supabase Postgres (RLS)	To write to you the right way and, soon, to build a glossary with the terms of your field. You can erase them whenever you want, in the same place.
`usage_logs` (user_id, action, timestamp)	On every cloud transcription/translation	Supabase Postgres (RLS)	Enforce the plan's daily limits. No audio/text content; counters only.
Reminder text + time (Pro only, opt-in)	When a Pro user creates a reminder	Supabase Postgres (RLS)	Sync reminders across the user's own devices so they fire everywhere. Stored under the authenticated user and readable only by them.
Message schedule text, recipient and time	When you create/edit a scheduled message (signed-in user)	Supabase Postgres (RLS)	Sync across your own devices and let you manage them in the dashboard. Stored under the authenticated user and readable only by you. After the send, the text is removed: only a summary remains (content type, file name, recipient and date).
Anonymous diagnostics (opt-in, off by default)	Only if the user enables "Anonymous diagnostics"	Supabase Edge Function	Report technical errors and WhatsApp DOM changes to improve reliability — including audio-loading failures (only the technical phase, e.g. "evicted"/"all_phases", plus version and browser). No message content, contact names, phone numbers, or message identifiers.
`user_settings.usa_chave_propria` (user_id, yes/no, date)	When you save or remove your own key (BYOK), and on sign-in	Supabase Postgres (RLS)	To know how many accounts use their own key, so we can decide whether the feature deserves investment. It is a yes/no: the key never leaves your browser, and we do not record which provider, how many times you used it, or what was transcribed.
`active_sessions` (user_id, browser identifier, label, timestamp)	On sign-in and when using account-based features	Supabase Postgres (RLS)	Maintain one active browser per account (Section 4.1 of the Terms), preventing credential sharing. See section 2.5.

E-mails we send: besides your account notices (payment, security, changes to the Terms), the ZapVox Team sends usage tips in the first days and a monthly bulletin with what changed — on the basis of the legitimate interest (LGPD, art. 7, IX) of someone who created an account in knowing what is already theirs. Every one of these e-mails carries a one-click unsubscribe link, by category or for everything, and the switch is also under My account → Team e-mails. Whoever unsubscribes receives nothing beyond what concerns their own account. We never sell or hand your e-mail to anyone.

        Note on usage_logs: We record only that an operation happened (to enforce your plan's daily limit), not what was transcribed. No audio, text, contact name or message identifier is persisted on the server. The table is protected by Supabase Row-Level Security — you can only read your own records.
      

        Important: The extension does not access your microphone—it only reads audio already received in WhatsApp Web. Recovery of deleted messages has been discontinued — the extension does not store that type of data. With BYOK (your own API key), audio goes directly from your browser to Groq/OpenAI and never passes through our servers.
      

2.3 Local storage for extension functionality

To provide a smooth experience, ZapVox temporarily stores the following technical data in your own browser (never on our servers):

Contact name cache: Sender names are stored locally for up to 30 days for quick display in transcriptions. These names are the same ones already visible in your WhatsApp Web. They leave the browser in one situation only, described in the table in section 2.2: along with the audio during transcription, as a spelling hint — and you can turn that off in Settings.
Per-contact notes (v3.7.0+): Notes you write via the 📌 button live in chrome.storage.local (key zapvox_notes_v1), up to 500 characters per contact, and are never sent to any server. They go away when you delete the note, clear the extension's data or uninstall it.
Diagnostic logs: A circular buffer of up to 200 entries is maintained locally for debugging. Logs contain only anonymized name initials (e.g., "JS***"), never full names, and are not transmitted externally.
Console sanitization (v3.4.4+): All diagnostic logs printed to the browser console (visible via F12) have phone numbers sanitized to ***@ before being printed. This means that, even if you share a console screenshot with technical support, no WhatsApp phone number leaks.

        Full control: You can remove this data anytime by opening ZapVox Settings → "Clear privacy data" within the extension, or by clearing extension data at chrome://extensions.
      

2.4 Data we do NOT collect

Browsing history or activity outside WhatsApp Web
Contacts, photos, videos, or text messages from WhatsApp
Microphone, camera, or location data
Payment information (processed by Stripe; we have no access)
Fingerprinting data (GPU, detailed hardware, etc.) — the browser identifier in Section 2.5 is a random number, not derived from your hardware

2.5 Browser identifier (one active browser per account)

Context: the extension may be installed on several computers, but the account is active on one at a time — activating it on one computer automatically disconnects the previous access; simultaneous use requires the "Additional computer" add-on (see Terms, Section 4.1).

To ensure your account is used by a single person at a time — as set out in Section 4.1 of the Terms of Service — ZapVox needs to tell one browser from another. To that end:

What it is: a random identifier (UUID) generated by the extension itself on first run and stored in chrome.storage.local. It is accompanied by a generic, human-readable label (e.g., "Chrome · Windows"), derived only from browser type and operating system, so you can recognize on screen which browser is active.
What it is NOT: it is not fingerprinting. It is not derived from your hardware, GPU, fonts, screen resolution, or any signal that could track you across websites. It is a random number with no meaning outside ZapVox, and cannot identify you on any other service.
What we use it for: to know which browser is your account's active one, to display the switch notice with the "Use here" button, and to enforce the account usage rules (Terms, Section 4): the extension detects and records the browsers/computers connected to the account, and those records may support measures against abuse — such as limiting, suspending or blocking the account, at any time — in the event of credential sharing or simultaneous use that was not purchased. It is not used for advertising, behavioral profiling, or cross-site tracking.
Legal basis (LGPD): performance of the contract and legitimate interest in preventing fraud and abusive shared-account use, with minimal impact on your privacy.
Retention: we keep only the most recent record per user (one per account) — we do not build a device history. The record is deleted along with the account (Section 8), and the local identifier disappears when you uninstall the extension or clear its data.

3. API Keys (BYOK)

If you choose to use your own API keys (Bring Your Own Key — BYOK), they are:

Encrypted with AES-GCM 256-bit before being stored
Protected by PBKDF2 with 600,000 iterations and a unique random salt per user
Stored exclusively in your browser's chrome.storage.local
Never sent to our servers—they are used directly with Groq/OpenAI. Important: in BYOK mode, the audio/text goes from your browser straight to your key's provider, and usage is debited from the quota/credits of your own account with that provider (Groq has a free tier; OpenAI charges per use, paid by you directly to them). BYOK covers the AI features (transcription, translation, summary, sentiment); plan features are not changed
When your key starts being used: on the Free plan with Automatic mode, your key is used first. On Pro (and Trial/Annual/Biannual), our cloud is used first and, upon reaching 90% of that feature's daily limit, the extension switches to your key — from that point on the audio/text no longer passes through our servers and goes straight to your key's provider. With no key configured, nothing changes: the cloud is used up to the plan's limit and stops there.

4. Security

AES-GCM 256-bit encryption for BYOK keys
PBKDF2 with 600,000 iterations (OWASP standard) and unique random salt per user
Communication exclusively via HTTPS
JWT tokens with expiration and automatic refresh
Rate limiting on Edge Functions to prevent abuse
Payload validation on all messages (content script ↔ background)
Restrictive Content Security Policy in manifest.json

5. Browser Permissions

Permission	Why We Need It
`storage`	Save settings, transcription cache, and encrypted keys
`notifications`	Show notifications when transcription completes
`identity`	OAuth authentication for login (Google or Microsoft, via Supabase)
`alarms`	Keep the Service Worker active during long transcriptions and manage daily usage resets
`tabs`	Detect WhatsApp Web tabs to send results
`host: web.whatsapp.com`	Inject content script to add transcription buttons to audio messages

6. Your Rights

ZapVox is operated from Brazil and the processing of your personal data is governed by Brazil's Lei Geral de Proteção de Dados (LGPD, Law No. 13,709/2018). Under it you have the right to:

Access: Request what data we hold about you
Correction: Correct inaccurate data
Deletion: Erase your account and all your data yourself, instantly, from the dashboard (My account → Delete my account and my data), or by asking us by e-mail
Portability: Receive your data in a readable format
Revocation of consent: Disable auto-transcription anytime via the extension popup
Information about sharing: Understand how your data is shared with third parties

We answer requests of this kind from anyone, wherever you live — we do not require you to be in Brazil to ask for access, correction or deletion of your data.

To exercise any of these rights, contact us using the information in Section 12 — or, for deletion, use the dashboard button.

How dashboard deletion works (v3.13.3+): once you confirm by typing the safety word, deletion is immediate and final: we erase your account, name, e-mail, usage history, glossary, reminders, schedules and connected-computer records. If there is an active subscription it is ended at the same moment and no new charge is made — if ending it at Stripe fails, the account is not deleted and you see a notice to try again. The refund policy (Terms, §10.1) is applied first, automatically. What remains: only the payment records tax law requires us to keep (LGPD, art. 16, I), unlinked from you (account identifier and e-mail removed). Nothing that lives only on your computer passes through our servers — for that, see below.

How to clear local privacy data: Open ZapVox Extension → Settings → click "Clear privacy data". This removes diagnostic logs and contact name cache without affecting your settings or account.

How to clear all extension data: Go to chrome://extensions, click "Details" on ZapVox, then "Clear site data". This removes all cached transcriptions, settings, and BYOK keys.

7. International Data Transfers

Some data (settings, cached transcripts, reminders) stays in your browser, but audio transcription and other AI features rely on cloud processing, so your data may be transferred to and processed on servers located in the United States (through Supabase and Groq). We implement appropriate safeguards to protect your data during international transfers, including standard contractual clauses and encryption in transit via HTTPS. By using ZapVox, you consent to the transfer of your data to the United States for processing as described in this policy (LGPD, art. 33).

8. Third Parties and Subprocessors

Service	Purpose	Privacy Policy
Supabase	Authentication, Edge Functions (transcription/translation for Pro plan)	supabase.com/privacy
Groq	Cloud transcription, translation, summary (primary provider) + BYOK	groq.com/privacy
OpenAI	BYOK transcription (when user provides own key)	openai.com/privacy
Google Gemini	Translation and summary fallback (via Edge Function)	policies.google.com/privacy
Stripe	Payment processing (Pro plan)	stripe.com/privacy

9. Data Retention

Transcripts: they stay in your browser only, in a local cache that clears itself after 90 days, or sooner if the cache reaches 7 MB (in which case the oldest go first). The transcribed text is not stored on our servers — it is generated, returned to your screen and discarded. The audio is never stored either: it is processed and discarded immediately (practice verified and corrected on 13/08/2026).
Deleted messages: not applicable — discontinued feature, nothing is stored.
Contact name cache: 30 days (automatic, local)—removable via Settings
Diagnostic logs: Last 200 entries in circular buffer (local)—removable via Settings
Cascade telemetry (opt-in, local): Counters in chrome.storage.local — never transmitted. Resettable via the "Clear" button in the popup.
usage_logs (Supabase): Quota records (user_id, action, timestamp) are kept for 90 days for daily-limit auditing. No content, no PII beyond your account user_id.
JWT session: Stored in chrome.storage.local until logout or refresh-token expiration (~30 days). We chose local over session to avoid forcing a re-login every time Chrome opens — Supabase RLS is the primary defense layer.
Scheduled messages and reminders (Supabase): when you schedule a message, the text to be sent, the recipient and the date are stored on our servers until it goes out — that is what lets a schedule survive closing the browser and show up on your other computers. The moment the send is confirmed, the text, caption, link and file are removed from our servers; only a summary remains — what kind of content went out (text, photo, file and its name), to whom and when — which is what lets your other computers know that send already happened and not repeat it. That summary is deleted after 30 days; what you deleted is removed after 7 days (an undo window). A send that could not be confirmed is kept whole for up to 30 days so you can check and resend. You can delete any schedule at any time from the dashboard or from WhatsApp Web.
Conversations with the site assistant: questions asked to the Help Center assistant are recorded to measure whether it answers well and to curb abuse. Do not use that field for sensitive data.
Account data: Retained while the account is active; erased instantly when you delete the account from the dashboard (Section 6). Payment records remain anonymized for the tax-retention period.
BYOK keys: Until you remove them or clear extension data
Processed audio: Discarded immediately after transcription (never stored)

10. Minors

ZapVox is not intended for children under 13 years of age. We do not intentionally collect data from children. If you believe a minor has provided data, please contact us for immediate removal.

11. Chrome Web Store Limited Use Disclosure

ZapVox complies with the Chrome Web Store's Limited Use policy:

Permitted use: Data collected is used exclusively to provide and improve the features described in this policy (transcription, translation, summary).
No advertising: Your data is never used for personalized advertising, retargeting, or ad targeting.
No data sales: Your data is never sold, licensed, or shared with third parties for commercial purposes.
No human access: No employee or contractor accesses your personal data except in case of legal obligation or with your explicit consent.
Limited transfers: Data is only transferred to the subprocessors listed in Section 8, exclusively to provide extension features.

11-A. Distribution through extension stores (Chrome, Edge and Firefox)

ZapVox is distributed through the Chrome Web Store (Google), Microsoft Edge Add-ons and Firefox Add-ons (Mozilla). To publish and update the extension we send those stores the program package. Mozilla additionally requires the extension's source code, because the package is minified — this is a mandatory condition to publish there, applied to every extension in that situation, and the material is used only by Mozilla's review team, under confidentiality.

What the stores NEVER receive, under any circumstance:

Audio, transcriptions, translations, summaries or any content from your conversations.
Your name, e-mail, plan, usage history or any account data.
Your API keys (BYOK) — they stay encrypted in your browser only and never leave it, neither to us nor to third parties.
Your glossary, scheduled messages, reminders or settings.

What we send is the program, not your data. Source code describes how the extension works; it does not — and could not — contain any user's information. This section does not change any commitment in Sections 7 and 11: data transfers remain limited to the subprocessors listed there.

12. Changes to This Policy

We may update this policy periodically. Significant changes will be communicated via notification in the extension or in the changelog. The last updated date will always be visible at the top of this page.

13. Contact

For questions, data requests, or to exercise your rights:

Email: [email protected]

Developer: Diegomaier Nunes Neri

Related documents: Terms of Service

In-app feature notices

To suggest features you haven't used yet (and avoid repeating ones you already use), the extension looks only at which features you have already used — a signal that already exists in your account. These notices collect no new data and never use the content of your messages or audios. You can turn them off in Settings → What's new.

AI providers (transcription)

Audio sent for cloud transcription is processed by one of these subprocessors, always in this order, moving to the next one only when the previous one fails: Groq → Google (Gemini) → Cloudflare (Workers AI) → OpenAI. They process the audio to produce the text and do not store it for that purpose. In BYOK mode, processing happens directly at your own key's provider (Groq or OpenAI), without passing through us.

Two transparency corrections made on 16/08/2026, so that this page describes what the code actually does:

Cloudflare (Workers AI) is now listed here. It was already part of the cascade and receives the audio when Groq and Gemini fail — it was not listed, and it should have been.
The day-of-the-week rotation was removed from the product on 15/08/2026 and has therefore been removed from this text. The order above is fixed, every day, for every plan.

Google (Gemini) receives the audio when it is reached in the cascade — not only the already-transcribed text. See the table in section 2.2.