Scroll Guard

Flags AI slop, shilling, clickbait, bot spam, and scams in your X feed with TypeSafe Jev. Bring your own key.

Scroll Guard flags AI slop, shilling, clickbait, bot spam, and scams in your X feed as you scroll. Each flagged post gets a colored frame and a short verdict. Open the verdict to see why: for AI slop, Scroll Guard names the strongest AI-writing pattern it found, such as a binary contrast ("It's not X. It's Y.") or a fake-profound kicker ("Let that sink in."). Every flagged post also gets a Block button that opens X's own confirmation. Nothing is ever blocked automatically. WHAT IT FLAGS • AI slop: posts that read as AI-written • Clickbait: framing built for clicks, such as "JUST IN: Study reveals…" • Shilling: posts that read as undisclosed paid promotion of a token, project, or product • Bot spam: copy-paste posts, reply farming, link drops, and keyword stuffing. Text that one account repeats, or that several accounts post word for word, is caught on your device without an API call. • Scams: fake giveaways, phishing, and requests for wallet recovery phrases Plain news, personal updates, jokes, and excited reactions are not flagged. YOU CONTROL IT • Turn each type on or off. • Choose how strict it is: set the threshold from 20% to 99%. It starts at 50%. • See how many posts it checked and flagged in the last 24 hours. BRING YOUR OWN KEY Scroll Guard uses TypeSafe's Jev model. You paste your own Jev API key from console.typesafe.ai/keys, and TypeSafe bills your usage directly. There is no subscription and no account with us. A typical day of scrolling costs about 1–2 cents. PRIVATE BY DESIGN • Only the text of posts on or near your screen goes to TypeSafe for analysis, including the post a reply answers, quoted posts, and the photo descriptions authors wrote. Photos themselves are never sent. • Your key stays on your device, in storage that pages on X cannot read. • No analytics, no tracking, and no servers of ours. • Open source: https://github.com/tameravci/scroll-guard A NOTE ON ACCURACY Scroll Guard asks Jev to judge each post against a written rubric. It is not a trained AI-text detector, and a verdict is evidence, not proof of who wrote a post. People write these patterns too. Scroll Guard is not affiliated with X Corp. or TypeSafe.