Privacy Policy — VexAI
VexAI is a browser extension that translates web pages, images, comics and documents, answers questions about the page you are reading, and summarizes YouTube videos. This policy explains exactly what data the extension handles, where it goes, how long it is kept, and what control you have over it.
We do not sell your data. We do not use it for advertising. The extension contains no analytics, tracking pixels, or telemetry about how you use it. The only thing it reports on its own is an anonymous install count, with no identifier attached (see "Install and uninstall counts" in §4).
1. Summary
| Do we sell or share data with data brokers? | No. |
| Do we use your data for advertising? | No. |
| Is there analytics or usage tracking in the extension? | Almost none — and never of your browsing. We never record the pages you visit, their web addresses or their content. For the first 24 hours after you install VexAI, the extension keeps a short diagnostic of how you use VexAI's own features — which button or shortcut you used, on what type of page (for example a comic page, a PDF, or a page Chrome blocks extensions on), and whether it worked — so we can understand why many new users remove it within minutes. This stops automatically after the first day. The exact data is listed in §4 ("New-install diagnostics"). Separately, right after it is installed it sends one anonymous message (extension version and interface language only) so we can count installs. |
| Do we read your browsing history? | No. We never collect a list of sites you visit, nor the address or content of any page. |
| Is an account required? | No. Web page translation, the selection popup and local OCR work without one; AI features require signing in with Google. |
| Is my content used to train AI models? | Not by us. AI providers handle your requests under their own data policies (§5). |
2. What the extension does not do
These are worth stating plainly, because the extension requests broad permissions:
- It does not run in the background collecting the pages you visit. Page content is read only at the moment you invoke a feature on that page (for example, you click "Translate" or turn on page reading in the Page Assistant).
- It does not transmit your cookies, passwords, or form data to us or to any third party. See §7 for the one narrow use of cookies.
- It does not record your keystrokes. The only keyboard listener that runs on pages reacts to the extension's own shortcuts (Alt+Q / Alt+W / Alt+S) and ignores every other key; nothing is stored or sent.
- It does not contain analytics, crash reporting, session recording, fingerprinting, or advertising code. The one-time install count (§4) carries no identifier.
- It does not scan, index, or bulk-read your Google Drive, Gmail, Sheets or Docs. See §8.
3. Account data
Signing in is optional; it is required only for AI-powered features and for purchasing credits.
When you sign in with Google, we receive from Google and store on our servers:
- your Google account ID
- your email address
- your display name
- your profile picture URL
We store this in our database (hosted on Supabase) together with your credit balance. We use it to identify your account, apply your credit balance, and contact you about your purchases. When you sign in for the first time we also send you a single welcome email; we do not send newsletters or marketing email. We do not receive or store your Google password.
When you create an account with email and password instead, we store your email address, a display name taken from the part of the address before the "@", and your password only as a salted one-way hash (scrypt) — we cannot see or recover the password itself. To confirm the address is yours, we email you a 6-digit code when you sign up, and again only if you ask to reset your password; signing in with your password sends no email. Codes are stored only as a keyed hash, expire after 15 minutes, and their records are deleted automatically one day after they expire. Accounts created this way receive no welcome email. If you later sign in with Google using the same address, both sign-in methods open the same account.
Your session token and profile are also stored locally in your browser (chrome.storage.local) so you stay signed in.
Free-credit abuse prevention. New accounts receive free credits once per device. To enforce this, the extension creates a random installation ID the first time you sign in (or first use a free feature without signing in, see below) and keeps it in chrome.storage.local; it is not derived from your hardware or browser characteristics. At sign-in our server receives this ID and the IP address of the request, and stores only a keyed hash (HMAC-SHA256) of each — never the raw values — together with whether free credits were granted. The hashed IP address is deleted after 30 days. The hashed installation ID is kept to stop the same device claiming free credits again, including after the account is deleted. Neither is used for anything else.
Daily free allowances. Some features have a free daily allowance (for example, a number of free comic pages per day, or free-model requests in Page Assistant). To count it we store your account ID, the date and the count. These counters are deleted after 7 days.
If you use these free features without signing in, the extension sends the random installation ID described above with the request instead of an account. We count the allowance against a keyed hash (HMAC-SHA256) of that ID and, to stop one network from generating unlimited new IDs, a keyed hash of the request's IP address — never the raw values — together with the date and the count. These counters are also deleted after 7 days. The content you send (the page text or comic image) is processed exactly as described for each feature in §4, and nothing links it to the hashed counters.
4. Data sent when you use a feature
Nothing below is sent unless you actively invoke the feature. "Our servers" means our backend at api.vexai.pyktools.com and our own OCR/detection servers; the providers named in each row are described in §5.
| Feature you invoke | What is sent, and to whom | Retained |
|---|---|---|
| Web page translation and the selected-text popup | The text segments to translate go directly from your browser to Google Translate. For a single English word, that word is also sent to the free dictionary service dictionaryapi.dev. Nothing goes to our servers. | Not retained by us |
| Page Assistant (chat) | Your message, the conversation history, your local date, time and time zone (so answers about "today" are correct), and — only while the 📖 "Read page content" toggle is on — the current page's URL, title and extracted text; plus any screenshot or image you attach. So the AI can still answer follow-up questions about them, the most recent images you attached in the conversation (up to 3, or 2 with the Free model) are sent again with your next messages; older ones are dropped. The toggle starts off every time you open the panel and resets when you navigate to another page, so page content is never sent until you switch it on for that page. Sent to our servers, then to the AI provider of the model you selected. If you enable Web search, the search query the AI writes is sent to a web search provider. | Conversation stored so you can resume it; you can delete any conversation from Settings → Page Assistant → chat history. Auto-deleted 60 days after its last update. |
| Comic / image translation — Free mode | The image goes to our OCR server. When the source language is set to Auto (the default), the image is also sent to Alibaba Cloud's AI model so it can tell which language the text is in, together with the language the page declares about itself. The recognised text is then translated either directly from your browser by Google Translate or on our servers by an AI provider. | Image not stored. The detected language is remembered so that other images — and other readers — of the same page skip detection: we store only a one-way hash (SHA-256) of the page address (including its query string, without the part after #) together with the language code. The address itself is never stored, and the record is not linked to you, your device or any account. Kept for up to 1 year, then deleted automatically; deleted earlier if a later image shows the stored language was wrong. Recognised and translated text may be cached — see §9. |
| Comic / image translation — Smart mode | The image, plus source and target language, go to our servers and then to Alibaba Cloud's AI model, which reads and translates it. | Image not stored. Translated text may be cached — see §9. |
| Snip & translate (OCR) | In the default mode, text is recognised on your device and the recognised text is translated directly from your browser by Google Translate. Other modes send the selected screen region to our OCR server or to an AI provider. | Not retained |
| Advanced / document translation | The text or document you submitted, sent to our servers and an AI provider (basic mode translates directly from your browser with Google Translate). | Not retained |
| AI content detector | The text or image you submitted, sent to our servers. Detection uses models hosted on our own servers and/or an AI provider. Text not in English may first be translated directly from your browser by Google Translate. | Not retained |
| Safe link — "Check link with AI" | Only the single link URL you selected from the right-click menu. Our server then fetches that page itself, without any of your cookies or credentials, and analyses its content with an AI provider. We never receive the links you merely hover over, click, or visit. | Not retained |
| Writing assistant and natural rewrite | Your prompt and the text you are working on — including text you ask us to rewrite so it reads more naturally — sent to our servers and an AI provider. | Not retained |
| News search / deep dive | Your search keywords and language. Our server queries news and web search providers and fetches the article pages it finds, then summarises them with an AI provider. | Search history stored. Auto-deleted after 90 days. |
| YouTube video summary | The video ID, title, channel name and length, and the video's subtitle (transcript) text, which the extension reads from the YouTube player in your own browser. Sent to our servers, then to the AI provider of our free model (summaries always use the free model, whatever model you selected for Page Assistant). Choosing another language on the summary card sends the summary text to be translated the same way. You can turn the card off, or make it summarize only when you click: Settings → YouTube. | Not stored on our servers. The finished summary is kept in your browser's local storage (the 50 most recent videos) so reopening a video does not run it again |
| YouTube dubbing (extension versions up to 1.0.3 only) | The video ID and the subtitle text to translate and voice, sent to our servers; the translated lines are voiced by a text-to-speech provider (Google Cloud or Microsoft) using its standard synthetic voices — VexAI never clones or imitates a real person's voice. If you choose the free voice, lines are sent directly from your browser to Google's text-to-speech service, or spoken by your device's own voices without any network request. | Generated audio is cached in server memory, not linked to your account, and cleared when the server restarts |
| Vibes (custom effects) | The effect you create: its name, settings, and either an emoji, an image you upload, or an image generated by Alibaba Cloud's AI model from the text prompt you write. Before generating, the prompt is checked by our own filters, by an AI model and by our payment provider Creem's moderation service, and refused if it asks for a real, identifiable person, sexual content or gore. There is no face swap and no way to upload a photo of a person for the AI to transform. | Stored until you delete the effect. The first few words of an AI prompt appear in your credit history, which is deleted after 90 days |
| Feedback form | Your message, the optional reply-to email you enter, and any screenshots you attach. Delivered to us by email (sent through our email provider, Resend); not stored in our database. | Kept in our mailbox for as long as needed to handle it |
| Uninstall survey (optional) | After you remove the extension, Chrome opens a short survey on our website (§11). Only if you choose to send it: the reasons you tick, your message, the optional reply-to email you enter, plus the extension version and interface language. Delivered to us by email (Resend); not stored in our database and not linked to any account. | Kept in our mailbox for as long as needed to handle it |
| New-install diagnostics (first 24 hours after install) | To understand why some people remove VexAI within minutes of installing it, for the first day the extension records how you use its own features, so we can fix buttons that silently do nothing. We record: a random install code (a throw-away number, not linked to your account, email or device ID); which VexAI button or keyboard shortcut you used and whether it worked or failed; the type of page you used it on — such as a comic/image page, a PDF, a YouTube page, a page Chrome blocks extensions on, or an ordinary web page (never the address, title or content of the page); which of our own backend requests the action triggered and the result code; and, worked out by our server from your IP address and then discarded, your country only. It records nothing about the pages you visit, nothing you type, and no account, device ID or IP address. It stops on its own after the first 24 hours. | Auto-deleted after 30 days |
| Install and uninstall counts | Once, right after you install the extension, it sends our server the extension version and your interface language. When you uninstall it and Chrome opens our survey page (§11), that page sends the version, interface language and the day the extension was installed, whether or not you answer the survey. No account, installation ID, device ID or IP address is stored with these counts; they only tell us how many people install and remove VexAI each day and how long they kept it. | Auto-deleted after 400 days |
| Credits and purchases | We keep a record of every credit change on your account (free credits, usage, refunds). Paid credit packs are sold through Creem (§5): you enter your billing and card details on Creem's checkout page — we never see your card. From Creem we receive the order ID, the pack bought, the amount paid and the email used at checkout, so we can add the credits and handle refunds. | Credit usage records auto-deleted after 90 days; purchase records kept 2 years (needed for refunds and disputes) |
Retention above is enforced automatically by scheduled jobs in our database.
5. Third parties who process data for us
| Provider | Role | What they receive |
|---|---|---|
| Supabase | Database hosting | Account record, chat sessions, search history, credit transactions, custom effects |
| Alibaba Cloud (Qwen models) | Default AI provider for translation, comic reading, OCR assistance, detection, writing and image generation | The content of the specific request you made (text, image, page extract, prompt) |
| OpenRouter, and through it the maker of the model you pick — Google, Anthropic, OpenAI, DeepSeek, Xiaomi, MiniMax, NVIDIA, inclusionAI, Mistral AI, Alibaba Cloud (Qwen) | AI answers in Page Assistant when you select one of these models or use the Free model | The conversation and any page content or image you chose to include |
| Sign-in; Google Translate; Cloud text-to-speech; Sheets/Docs API (§8) | OAuth identity check; text you translate; subtitle lines to voice | |
| Resend | Email delivery: sign-up and password-reset codes, the welcome email, and forwarding feedback messages to our inbox | Your email address and the code (sign-up / reset); your email address and name (welcome email); feedback messages |
| Creem | Payment processing for credit packs, as reseller (Merchant of Record) | Your email and order details at checkout (card details go directly to Creem — we never see them) |
| Microsoft | Neural text-to-speech voices for YouTube dubbing (extension versions up to 1.0.3) | Subtitle lines to voice |
| Web search providers — Serper, Google Programmable Search, Tavily, DuckDuckGo | Web and news search for Page Assistant and News | The search query only — never your identity |
| dictionaryapi.dev | Dictionary definitions in the selection popup | The single English word you selected |
| Sentry | Error reporting on our servers (not in the extension) | Technical error details: error message, the feature involved, your account ID. Images, tokens, cookies and authorization headers are stripped before sending. |
Free model in Page Assistant. The optional "Free model" costs you no credits, but it is not a no-cost model: we pay for it. Text questions currently run inclusionAI's Ling model; questions that include an image run Alibaba's Qwen model; Mistral AI's Mistral Small is the backup for both. All of them are reached through OpenRouter — the same way as the paid models, under each provider's own data policy. You can use the Free model without signing in, within the same limits (see §3).
OCR for Free-mode comics and the AI-image detection models run on servers we operate ourselves; that content is not passed to a third party for those steps.
6. Why the extension needs each permission
| Permission | Why it is needed |
|---|---|
<all_urls> / activeTab / scripting | To read and translate the page you are currently on, and to draw the translation overlay and side panel. Content is read only when you invoke a feature. |
storage | To save your settings, notes, and sign-in session |
tabs | To know which tab a feature was invoked on, and to open the settings page |
contextMenus | The right-click "VexAI" submenu on selected text (Translate, Summarize, Explain, Reply), and — when Safe link check is enabled — the "Check link with AI" entry shown on links |
identity | Google sign-in |
cookies | Two narrow uses only: (a) when downloading an image that is protected against hotlinking, we re-attach the cookie you already have for that image's own site so the download succeeds; (b) in extension versions up to 1.0.3 only, we attached your youtube.com cookie when fetching YouTube subtitles for dubbing. In both cases the cookie is sent only back to the site it belongs to. Cookies are never sent to VexAI or to any third party. |
offscreen | Runs the on-device OCR engine used by Snip & translate, which needs a worker in Manifest V3 |
declarativeNetRequestWithHostAccess | To set the correct Referer when fetching images from sites that block hotlinking |
Host access to translate.googleapis.com, www.googleapis.com, accounts.google.com | Google Translate requests made from your browser, Google sign-in, and the Google Sheets / Docs skills |
7. Google Sheets and Google Docs access
Signing in to VexAI asks only for your basic Google identity (openid, email, profile). Two further scopes exist, and they are requested separately and only at the moment you first open the Google Sheets or Google Docs skill — never during normal sign-in:
.../auth/spreadsheets— read and edit Google Sheets.../auth/documents— read and edit Google Docs
If you only use VexAI to translate and read, you are never asked to grant these at all.
We deliberately do not request any Google Drive scope. The extension cannot list, browse, or download files from your Drive — it can only act on the specific Sheet or Doc you have open at the time.
These scopes are used only by the Page Assistant's optional Google Sheets / Google Docs skills, and only when you have explicitly enabled that skill and are working on such a document. In that case:
- the relevant cell range or document content is sent to the AI model so it can answer or perform the edit you asked for;
- any edit is written directly from your browser to the Google API;
- we do not browse, list, index, or back up your Drive, and we do not store your Sheets or Docs content on our servers.
Every operation that writes to a Sheet or Doc shows a confirmation dialog first. That dialog lists the actual file ID, range and change taken straight from the request that is about to be sent — not just the AI's description of it — so you can check what will happen before allowing it.
VexAI's use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.
If you never enable these skills, these scopes are never exercised. You can revoke access at any time from Settings → Page Assistant → Google Sheets & Docs connection → Revoke, or at myaccount.google.com/permissions. Signing out of the extension also revokes it automatically.
### Confluence pages
The Page Assistant can also read and edit a Confluence page (Atlassian Cloud or a self-hosted Data Center site) — only the page you have open, and only after you turn on 📖 page reading on it. It requests no extra permission and no Atlassian account access: it calls Confluence's own REST API from that page, with the sign-in you already have in your browser, so it can do exactly what you can do on that page and nothing more.
- The page content (and the names of its attachments) is sent to the AI model so it can answer or prepare the edit you asked for. Files you attach in the chat for that page are kept in your browser until an edit uses them.
- Edits and file uploads go directly from your browser to your Confluence site — they never pass through our servers, and we do not store your Confluence content.
- Every edit opens a review listing each change (before → after). Only the changes you keep ticked are saved, as one new page version that you can compare or restore in Confluence's Page History.
8. Shared caches
Image translations. To reduce cost and latency, translations of images are cached on our servers keyed by a hash of the image, and reused for anyone who later translates the identical image. The cache entry contains the extracted and translated text — it is not linked to your account or to any user identifier, and entries expire after 180 days. Pages you capture from the screen (rather than images loaded by the site) are never written to this cache.
Because the translated text of an image you process may be served to another user who processes the same image, do not use image translation on private or confidential images while this is enabled. You can turn it off: Settings → Comic → "Enable global translation cache". With it off, your translations are not written to the shared cache.
Dubbing audio (extension versions up to 1.0.3). Voiced subtitle lines are cached in server memory keyed by a hash of the voice and text, so a popular video is only voiced once. The cache holds no user identifier and is cleared whenever the server restarts.
9. Features that never leave your device
These run entirely locally and make no network requests with your content:
- Gmail pre-send check — the confirmation dialog before you send an email. Your email content is analysed in the page and never transmitted.
- Safe-link check (automatic warning) — the warning shown when you click a link with a suspicious URL structure is computed entirely in the page. No link you hover, click, or visit is transmitted. The separate "Check link with AI" action is not local — see §4.
- Quick Notes — stored in
chrome.storage.localon your machine only. - On-device OCR in Snip & translate — the image is processed in your browser. The first time you use a language, the extension downloads that language's recognition data from
tessdata.projectnaptha.com; this download contains nothing about you or your image. - All extension settings and panel layout state.
10. Data stored on your own device
| Location | Contents |
|---|---|
chrome.storage.sync (synced by Chrome to your Google account) | Settings: languages, theme, enabled sites, translation mode |
chrome.storage.local (this device only) | Sign-in token and profile, a random installation ID (§3), quick notes, translation cache, panel state, and a count of how many times a feature worked for you — used only to decide when to show a one-time invitation to rate VexAI on the Chrome Web Store; it is never sent anywhere |
Uninstalling the extension removes both. Removing your server-side account requires a request — see §12.
11. Our website
The website at vexai.pyktools.com is a static site with no sign-in, cookies or tracking, and no analytics apart from the anonymous uninstall count described below (and, when the extension is still within its first 24 hours, the uninstall page also sends the new-install diagnostics event — see §4). It is hosted on GitHub Pages; like any web host, GitHub may log the IP address and browser details of visitors for security purposes, under GitHub's own privacy statement. The demo video on the home page is a still image until you press play; only then is the video loaded from YouTube in its privacy-enhanced mode (youtube-nocookie.com), under Google's privacy policy.
When you uninstall the extension, Chrome opens our survey page (vexai.pyktools.com/uninstall/). Its address carries only the extension version, your interface language (so the page can show in your language) and the day the extension was installed. When the page opens it sends these three values to our server as an anonymous uninstall count (§4). Answering the survey is optional; nothing else is sent unless you press "Send feedback" (see the uninstall survey in §4).
12. Your rights and choices
- Access / export — Settings → Account → Export data copies your local settings and data.
- Delete a conversation — Settings → Page Assistant → chat history, at any time.
- Delete your server-side data — email support@pyktools.com from the address you signed up with. We will delete your account record, chat sessions, search history and custom effects, and confirm within 30 days. Purchase records are kept for the period in §4 where the law or payment disputes require it, and the hashed installation ID described in §3 is kept. (There is currently no self-service delete button; this is handled manually.)
- Delete local data — sign out, or uninstall the extension. Signing out keeps the installation ID (§3); uninstalling removes it.
- Withdraw Google access — see §7, or myaccount.google.com/permissions.
- Opt out of the shared image cache — see §8.
Depending on where you live you may have additional rights (access, rectification, erasure, portability, objection) under the GDPR or similar laws. Use the contact address above to exercise them.
13. Security
Traffic between the extension and our servers uses HTTPS. Access to your account data is protected by a signed session token, and database rows are isolated per user by row-level security. No system is perfectly secure; please report vulnerabilities to support@pyktools.com.
14. Children
VexAI is not directed to children under 13, and we do not knowingly collect data from them. If you believe a child has provided us data, contact us and we will remove it.
15. International transfers
Our infrastructure and AI providers operate in multiple regions, including outside your country of residence. By using the extension you understand your request content may be processed in those regions.
16. Changes to this policy
If we change how data is handled we will update this page and revise the "Last updated" date. Material changes will also be noted in the extension's release notes on the Chrome Web Store.
17. Contact
Questions, requests, or complaints: support@pyktools.com