An open-source demo that shows what an exam platform can and cannot detect from the browser during a 60-second session. No camera, no server, no install: just native browser signals, a notebook aesthetic and a pair of eyes that follow your cursor.
Tech-stack I used here
Astro
React
OpenCode
TypeScript
Tailwind
Cloudflare
Figma
Origin and purpose
A question worth testing: Do online English tests really detect translator or side-tool use? I wanted to check it from the browser side.
No tool covered it: Nothing bundled every detection signal in one free, fun place, so I built it.
A timed, honest demo: I Can See You shows in 60 seconds what exam platforms can and cannot detect, with a doodle identity and a mascot that follows your cursor.
Decisions
Browser signals only: No OS-level surveillance claims. The demo shows what it can see and what it cannot.
Client-side by design: No backend or database. sessionStorage moves results from /demo to /results.
Astro + React islands: Static shell and SEO in Astro, interactivity only where needed in React, with strict TypeScript throughout.
A playful identity: Cursor-tracking mascot eyes and a monochrome doodle look keep a spying concept approachable.
A 60-second session: Auto-starting and stoppable early, short enough to feel real and fast to demo.
Browser-only detection
Browser-only detection pipelineSix browser signals are grouped by confidence and passed through a client-side detection engine, an event store and a results builder before the session is handed to the results page through sessionStorage, all inside the browser sandbox with no backend, database, camera, microphone or OS access.Browser sandbox · client onlyBrowser signalsHighvisibility · blur / focusMediumfullscreen · mouseleavepasteLowdevtools heuristicDetectionEnginenormalizeopen / closeEventStoreopen + closedincidentsResultsBuildermetricsattention samplesResults pagesessionStoragehandoffNo backend · no database · no camera · no microphone · no OS accessBrowser-only detection pipelineSix browser signals are grouped by confidence and passed through a client-side detection engine, an event store and a results builder before the session is handed to the results page through sessionStorage, all inside the browser sandbox with no backend, database, camera, microphone or OS access.Browser sandbox · client onlyBrowser signalsHighvisibility · blur / focusMediumfullscreen · mouseleave · pasteLowdevtools heuristicDetectionEnginenormalizeopen / closeEventStoreopen + closedincidentsResultsBuildermetricsattention samplesResults pagesessionStoragehandoffNo backend · no databaseno camera · no microphoneno OS access
Results
Live, fast and open source: On icanseeyou.dev, the Lighthouse mobile audit scores 94 for performance, 100 for accessibility, 100 for best practices and 100 for SEO, plus 2/2 for agentic browsing. The site is deployed on Cloudflare Pages, MIT licensed and open source on GitHub.
Real traffic, bots excluded: Cloudflare Analytics recorded 280 visits between Sep 19 and Oct 19, 2026: 140 from Google, 40 from Bing and 100 direct, with visitors from 19 countries.
Spanish is the main entry point: The Spanish landing page received 190 visits, ahead of the English landing page with 80. The bilingual bet paid off.
A complete session in 60 seconds: The demo auto-starts, tracks incidents live and ends on a dashboard with incident count, distracted time, attention score and a session timeline. Results can be shared as a 1080x1920 PNG through the Web Share API, with a download fallback.
Private by architecture: No backend, no database, no camera and no microphone. Session data stays local and is cleared when the results page loads or the tab closes; optional clipboard and multi-screen permissions are requested only when needed and degrade gracefully when unsupported.