Privacy

What Lexora processes

Lexora processes selected text and bounded reading context to provide translation, reading assistance, and pronunciation. Depending on the feature, context can include nearby text, a page or section title, and relevant link information. It does not collect a passive browsing history or upload whole pages for translation.

In the default Confirm mode, you select text and click the Lexora chip to open the panel. If you choose automatic activation, selecting text opens the panel and starts enabled features automatically. Enabled Wikipedia lookup, automatic pronunciation, and YouTube DualSubs can make requests within their workflows without a separate click for every request.

Three translation modes

Your translation mode controls the translation route. Other enabled features have their own network behavior.

• On-device Translation: Translation runs locally in the Reading Panel using Chrome’s Translator API. Translation text is not sent to a remote translation provider, and errors do not trigger remote translation fallback. Chrome may download browser-managed language packs when you prepare a supported language pair. This mode does not make Wikipedia, pronunciation, or YouTube offline.
• Google Translate: Selected text is sent directly to Google’s translation service (translate.googleapis.com), along with source and target language information. This mode does not require or load Cloud AI keys.
• Cloud AI: Optional translation sends selected text and relevant bounded context, such as nearby text, page or section titles, and glossary hints, directly to the selected OpenAI, Google Gemini, or Anthropic provider over HTTPS using your own API key. Only the selected provider and its key in the chosen storage mode are used. A missing key produces a configuration error; Cloud AI does not silently switch providers or fall back to Google Translate.

Google Translate is the default translation mode. On-device and Google translation do not use Cloud AI credentials. Test connection is a separate action that sends fixed test text to the chosen cloud provider using your key. Lexora does not operate a server or proxy for these requests.

Other reading services

• Wikipedia and Wikidata: When Wikipedia is enabled, selection-derived search terms, article titles, or relevant Wikipedia links can be sent to Wikipedia to retrieve reference context. Wikidata receives entity identifiers (QIDs) to find articles in other languages. Resolution can make several requests automatically; some lookup or excerpt translation paths also use Google Translate independently of the selected translation mode.
• Pronunciation and Japanese reading: Audio playback can send source or translated text to Google TTS (translate.googleapis.com or translate.google.com). Automatic pronunciation can do this when enabled. A browser speech fallback may use a local or remote voice, depending on the browser and voice. Japanese reading fallback can send text to Google’s service.
• Local reading aids: Bundled Japanese tokenization, furigana, and English IPA lookup run locally. Panel rendering and subtitle synchronization also run locally.

YouTube DualSubs

On YouTube, DualSubs reads video and caption-track information and fetches source and translated captions from YouTube’s timedtext service. Requests include video/track identifiers and language choices and can include your existing YouTube session cookies. DualSubs is enabled by default and can load captions automatically on supported video pages; you can turn it off in settings or the player controls. Caption translation uses YouTube’s service independently of the Reading Panel translation mode and Cloud AI keys. Caption cues are held in the page session for display and synchronization, rather than uploaded to Lexora.

API keys, storage, and retention

Settings and DualSubs preferences are stored in chrome.storage.local in your browser profile. User-provided Cloud AI keys are stored separately in chrome.storage.local (persistent, the default) or chrome.storage.session (temporary for the browser session), according to your choice. Lexora does not sync these through chrome.storage.sync. Keys authenticate requests directly to the relevant provider; they are not sent to Lexora. You can clear saved keys in settings. Browser extension storage is not an encrypted password vault.

Notebook saves are explicit. Entries contain selected text, translation, language pair, selection type, source hostname, and saved date, plus optional explanation, a bounded context snippet, and review metadata. Entries remain in local extension storage until deleted or the extension is removed. You can delete individual entries or all entries, or export Markdown locally. Export excludes full page URLs, page titles, API keys, and diagnostics. Saving from incognito tabs is blocked. Exported files remain under your control and are not removed when you delete Notebook entries.

Reading results, translations, lookup data, and subtitle cues may be held in temporary in-memory caches or active UI state. Cache lifetimes vary by feature; Lexora does not promise immediate erasure after each request. On-device translation does not add a translation-result cache. External providers receive normal connection metadata such as your IP address and may retain requests under their own terms and account settings. Lexora cannot delete their records or promise zero retention or no model training by a provider.

What Lexora does not do

• Lexora does not sell user data or use it for advertising, credit decisions, or data brokerage.
• Lexora does not require a Lexora account to use the extension. Cloud providers may require their own accounts and billing.
• Lexora does not include analytics or remote telemetry collection. Local console diagnostics can contain operational metadata and errors; they are not sent to a Lexora telemetry service.
• Lexora does not operate a backend that collects your selected text, API keys, Notebook, or browsing history.

Privacy controls and redaction

Use settings to choose the translation mode and control Wikipedia, automatic activation, automatic pronunciation, and YouTube DualSubs.

• Choose On-device Translation to keep translation text local, and review other enabled features separately.
• Choose session-only key storage or clear saved keys; delete Notebook entries when no longer needed.
• Redaction settings replace recognized email addresses, IPv4-shaped addresses, and certain known secret-key patterns in supported outbound translation, Cloud AI context, Wikipedia query, and Japanese reading paths. Email and secret-pattern redaction are on by default; IP redaction is off by default.

Redaction is pattern-based, not a guarantee of anonymization: it does not recognize every secret or sensitive detail. It does not apply universally to all network paths, including pronunciation audio, direct article links, and YouTube caption requests. Avoid selecting confidential text for external processing.

Chrome Web Store Limited Use

Lexora’s use and transfer of user data, including information received from Google APIs, adheres to the Chrome Web Store User Data Policy, including its Limited Use requirements. Data is used only to provide the disclosed reading, translation, pronunciation, subtitle, and Notebook features. Transfers to providers described here are limited to what those features need.

Chrome Web Store User Data Policy — Limited Use

This website

lexora.sh is a static site hosted on GitHub Pages. It does not set tracking cookies, collect form submissions, or run analytics. Your language and theme preferences are stored locally in your browser. Hosting and external service providers may process normal request metadata under their own policies. Contact is by email link; information you choose to send by email is handled to respond to your inquiry.

Contact

Privacy questions or data-handling concerns: [email protected]