Chrome Web Store

Transight AI

3825269c-b209-535c-ac57-6c0305c12fa3 | Store v0.1.6
0/ 100
MINIMAL risk
-52 since v0.1.1
AI review verdict
No high-risk signal observed

Based on the RiskyPlugins AI security review of the observed evidence. The review covers version 0.1.5, not the analysed version 0.1.6.

Analysis record

Analysed
1 month ago 22 October 2026
Version
v0.1.6
Artifact
SHA256 99b805a3…5b84da
Detail
Full list

No findings

The analyzer reported nothing for this version.

Publisher Evidence

Limited evidence

dev--cd7de494-a58c-8d6f-0394-5c8410f79075

Publisher identity, store signals, distribution reach, and warning signals used for context. Treat this as supporting evidence, not a clean bill of health.

Chrome does not expose the same publisher verification data as IDE stores, so this score is deliberately conservative.

29
Publisher domain
No domain
Missing
Store verification signal
Limited signal
Limited
Extension portfolio
2
Portfolio

12 evidence rows available.

No findings

The analyzer reported nothing for this version.

AI Security Report

AI Security Review

Evidence context: threat category none; evidence quality moderate.

Transight AI is a small Chrome extension (version 0.1.5, three users) that translates selected text through an AI service the user configures. The listing says you highlight text, right-click, and the extension sends it to your own endpoint. The scan produced four network findings, all of them fetch calls in the extension's own source.

The locations line up with that description. src/shared/i18n.js:39 sits in the localization loader, which is where extensions call fetch to pull in a string table for the interface. src/screenshot/screenshot.js:28 and src/shared/screenshot.js:77 sit in the capture path, the code that reads an image or a captured tab so there is something to translate. src/shared/selection-background.js:38 handles the right-click selection flow the listing advertises. Four files, four ordinary uses of a browser API that a translation tool needs.

Nothing else was found. No malware signatures matched. No obfuscation was detected. No indicators of compromise, which means no unknown or suspicious domains were pulled out of the code. No hardcoded secrets. No dependency or tool-poisoning findings. For a codebase that ships an i18n layer and a screenshot pipeline, that is a thin record.

A fetch finding records that the API is called. It does not record a destination. The manifest capture for this build returned no declared permissions, no host permissions and no extracted network endpoints, so the evidence cannot say which hosts these calls reach. For an extension whose selling point is "your own AI service", a user-configured destination is the expected design, and a translation tool that never made a network call would not work at all.

The strongest argument against this reading is that software which captures screenshots and sends text over the network is a plausible path for moving page content off a device, and with three installs there is no track record to lean on. That is a fair question about capability. It is not what the findings show. The findings are the API call, not a hostile host, and no file in the evidence pairs page content with an unidentified domain.

What would change this: a manifest declaring host permissions for a domain that is not the endpoint the user chose, or an extracted endpoint belonging to neither the user's AI provider nor the browser itself. Neither is present in what we captured. The gap worth closing is the empty manifest, which leaves the permission picture unverified rather than clean.

Key Reasons

  • All four findings are generic fetch-API detections in source files, with no destination recorded
  • Zero malware signatures, obfuscation, IoC, secret or dependency findings in the bundle
  • File paths (i18n.js, screenshot.js, selection-background.js) match the declared translate-on-selection behavior
  • Manifest capture returned no permissions, host permissions or network endpoints, so no suspicious host appears anywhere

False Positive Considerations

  • NET-FETCH rules flag any use of the fetch API regardless of where it points
  • Permission and endpoint extraction returned empty for this build, so no hosts were resolved

Reviewed 2026-10-17 · version 0.1.5; recommended action: suppress false positive; model confidence 72%.

Source Code Viewer

Source Code Not Available

Source code is not available for this version of the extension.

About This Extension

Transight AI — Across languages. Closer in meaning. Translate selected text, typed text, pasted images, or an entire webpage using your own AI service. Open the toolbar popup, click the icon beside a selection, or use the right-click translation command. AI WEBPAGE TRANSLATION — STAY ON THE PAGE • Choose AI webpage translation in the popup or a selection/right-click panel. Review the provider, first configured model, target language, privacy notice and possible costs, then confirm before page text is sent. • Ordinary rendered text is translated in place. Page elements, links and event handlers are retained; translated text can change line wrapping and element heights. • Stop, compare original/translation, restore and end the session, or collapse the toolbar. Newly revealed or added text is processed during the authorized session. A URL change ends the session; continuing after a round limit needs confirmation. • Forms, editors, code, hidden text and no-translate regions are skipped. Other page text can still contain sensitive information. Iframe content, Shadow DOM, PDFs and image/canvas text are not supported. Uses a dedicated structured prompt rather than the custom text-translation prompt. SIMPLE OR FULL • Full mode compares up to 5 models in separate cards. Copy, cancel, or retry failed results independently. • Simple mode uses only the first model, with editable source text and source/target selectors. It translates automatically after a short typing pause or language change, without a Translate button. • Switch modes from the layout icon or Settings. Mode and target language preferences stay synchronized across popup, panels and Settings. • Select text to show a compact icon near its end. Selection alone sends nothing. Click to translate, pin the panel, or drag its top bar. • Copy and speech controls are available in both modes. Speech uses an installed local Chinese voice for all languages; pronunciation quality varies. No automatic playback. IMAGES AND SETTINGS • Paste up to 5 clipboard images in Full mode; thumbnails scroll horizontally and can be removed individually. Requires a model that accepts images. Previewing alone does not upload images. Use a system screenshot tool, then paste. • Enable/disable automatic selection controls, choose translation style, fetch a model list or enter model IDs, and customize the text-translation system prompt. Optional page context for text/selection prompts uses available title/description metadata, not an AI-generated whole-page summary. • Five interface languages: English, Simplified Chinese, Traditional Chinese, Japanese and Korean. Browser language is followed by default with English fallback. Ten target languages are supported. SEPARATE GOOGLE SHORTCUTS • Google whole-page translation opens the original HTTP/HTTPS page URL with the current target language in a new Google tab. It does not modify the original page; private or login-only sites may not work. • Google free text translation opens up to 5,000 source characters with the chosen languages in a new tab. It can be disabled in Settings. Neither entry requires an AI key. Results stay on Google. • Only clicking sends text or a URL to Google. No AI credentials or images are included. URLs/text may appear in browser history; Google policies and availability apply. SETUP AND PRIVACY Enter a Chat Completions-compatible Base URL, API key if needed, and model IDs. To persist a key, create a local unlock password of at least 6 characters. Keys are encrypted locally using AES-256-GCM with a PBKDF2-derived key; unlocked keys are held in restricted session storage. Unlock inside the popup/panel when needed. Passwords are not sent to AI providers. AI requests send submitted text, images and applicable prompt metadata to your configured endpoint. After explicit AI webpage confirmation, ordinary page text, including newly revealed content, is sent in batches; whole HTML/DOM, page URL and cookies are not included in those requests. Original/translated page text remains in memory, not persistent history. API authentication is sent to the configured provider. Model discovery and connection testing also contact that provider. Multiple models and automatic retranslation can increase usage costs. No service, key, credits or subscription is included; there is no built-in analytics or persistent translation history. Protected browser pages cannot host on-page controls.

Frequently Asked Questions