GemaShield

Created by team kaizenv2 on May 18, 2026
AI Agents with Google AI StudioAgent Security & AI Governance - VeeaData & Intelligence

A Chrome extension that summarises any news article, flags structural misinformation cues, warns about sensitive content, and reads it all aloud, entirely on your device, with zero API keys and zero data leaving your machine. The problem Reading news online is exhausting in three different ways: Volume. There's more written every hour than anyone can keep up with. Quality. Misleading framing, missing sources, and loaded language slip past even attentive readers, and "is this true?" tools usually mean sending the page to a cloud service. Access. People with low vision, dyslexia, or sheer reading fatigue need audio, but the built-in browser readers are clunky and inconsistent. Existing solutions hit one of those three. Most of them also send the article you're reading, plus a record of what you read, to a server somewhere. What GemaShield does One click on the toolbar icon and the extension extracts the article from the active tab, then runs three on-device passes in parallel through Chrome's built-in Gemini Nano: Gist. A streamed, two-to-three-sentence summary so you can decide whether to read the rest. Misinformation check. A structured analysis of structural red flags: unverified claims, missing or anonymous sources, emotional framing, logical fallacies, conflicts of interest. Each flag comes with a one-line explanation. It does not pretend to fact-check (the model has no internet); it surfaces patterns that earn a closer read. Sensitive content warnings. Categorised heads-up about violence, hate, self-harm, distressing imagery, and similar, with a calibrated severity rating so calm news stays uncluttered. Then the analysis auto-plays through the system's text-to-speech voice. A two-row control bar gives play / pause / stop, speed (0.6×–1.6×), and a voice picker that lists every voice installed on the user's OS grouped by language. The choice persists via chrome.storage.local, and a "test" button previews the voice in its native language (16 supported)

Category tags: