At a glance
- The extension may use hosted processing while its verified browser model downloads, then prefer a hybrid local-model path.
- Reading history is off by default and is not collected unless you enable it.
- The website demo sends only the text you submit to the requested reading services.
- Optional hosted features disclose that text will leave the device before they run.
- VowelMarks does not sell personal information or use reading text for advertising.
The Chrome extension
The extension needs access to the page where you ask it to add reading assistance. It reads the relevant Persian text, applies the result to that page in your browser, and preserves the original letters. Browser and protected Chrome pages may not allow content scripts to run.
The standard extension silently downloads an integrity-checked Persian model pack after installation. Before that pack is ready, requested text is processed by the VowelMarks hosted engine. Once ready, the extension can run model prediction locally and send the source plus bounded structured predictions to VowelMarks for protected finalization. The standard launch version is not a fully offline product.
The website demo
Text pasted into the website demo is sent to VowelMarks’ server-side diacritization API and processed in memory. Normal engine text is not written to application tables or operational logs. The demo is rate-limited, including privacy-preserving network-address controls, to protect the service. Do not paste private, confidential, or identifying text.
After you request reading cues, the website may also start a word-origin/Persian-alternative lookup in the background so it is ready if you open that view. English translation and speech remain separate requests that begin only when you ask for them. A failed optional request does not change or remove the marked Persian result.
Completed website-demo results may stay in this browser tab's session storage so the compact homepage tool can continue in the full web tool without repeating work. This temporary cache is not synced to an account, is not reading history, and is cleared when the browser session ends.
Optional hosted features
Explanations, coaching, word help, quizzes, grading, and learning summaries use managed Gemma through VowelMarks and send the bounded text you submit to Google. Translation uses Google Cloud Translation, while Premium Voice uses Google speech. Do not submit sensitive or confidential text. VowelMarks does not store AI prompt or response content in server logs.
Accounts and authentication
Clerk provides email-code and Google sign-in, account recovery, and session management. VowelMarks receives a signed account identifier and basic account information; it does not receive or store your password. In the extension, session tokens stay in the background service worker and are not exposed to page scripts.
Synchronized learning history
Learning history begins only after you opt in. If you later sign in, consented history can synchronize automatically across devices even when Active Reading is off. Synchronized data may include a sanitized page title and hostname, a URL with credentials/query/fragment and token-like path portions removed, bounded excerpts and passages, contextual word readings, help actions, verified quiz attempts, self-reports, and derived learning patterns. VowelMarks does not retain unrestricted full-page copies for this purpose.
You can pause collection, exclude sites, export data, and delete cloud learning history. Incognito browsing is excluded. Deletion advances an account deletion marker so an older offline device cannot restore the deleted records when it reconnects.
Learning retention
Active accounts retain detailed learning evidence subject to record caps. When a cap is reached, older detail may be compacted into word, pronunciation, skill, mastery, and quiz aggregates. After 12 months without VowelMarks activity, VowelMarks plans to send a notice, wait 30 additional days, remove detailed page identities and passage text, and retain compact learning aggregates until you delete them or close the account. That inactivity deletion will not run until notice delivery is operational.
Subscriptions and payments
Polar is the merchant of record for VowelMarks subscriptions. Payment-card and billing details are entered on Polar’s hosted checkout and do not enter VowelMarks systems. VowelMarks stores the identifiers and verified subscription state needed to provide account access, handle refunds, and meet legal obligations. Legally required transaction records may remain after account deletion.
Infrastructure and service providers
Cloudflare processes API requests and stores VowelMarks account, quota, billing-cache, and consented learning data. Railway processes hosted and hybrid Persian reading requests in memory. Hugging Face serves public model files and does not receive the Persian page text through VowelMarks. Requested Google Translation or Gemini coach/speech features receive only the bounded content needed for that request. Polar Cost Insights receives cost metadata, not Persian text, prompts, URLs, audio, quiz answers, or learning history.
Access and teacher forms
Access forms collect your email address and any optional context you choose to provide. This information is used to manage access, contact you about the requested pilot, and understand the kinds of Persian material readers bring to the product.
Analytics
The website uses self-hosted Umami analytics when configured. It records page views and product events such as running the demo or clicking an access CTA. VowelMarks does not send pasted Persian text, email addresses, or free-form application answers to analytics.
Your choices
You can leave reading history off, avoid optional hosted features, pause or exclude learning collection, export/delete cloud learning data, clear local extension storage, manage your account, and request account deletion.
Contact
Questions or deletion requests can be sent tohello@vowelmarks.com.