Sharp

Filter your timeline with your own AI provider, word rules and per-author rules.

Sharp hides the posts you don't want on X (Twitter). You say what that is, in your own words. Type "no crypto shilling" or "hide engagement bait and rage bait" or "only keep software engineering". Sharp reads each post on your timeline, scores how sure it is that the post matches, and folds the ones that do behind a slim banner. One tap shows any of them again. HOW IT WORKS • You write the filter in plain language. No keyword lists to maintain. • A fast decision model scores every post from 0 to 100%. • You set two lines: where hiding starts, and how wide the "close call" margin is. Close calls stay one tap away behind a tinted banner, so nothing important disappears silently. • Undecided posts wait behind a placeholder, so the timeline never jumps around. YOUR KEY, YOUR BILL, YOUR DATA • There is no Sharp account and no Sharp server. Nothing is collected, tracked or sold. • You connect your own AI provider with your own API key: OpenRouter, Vercel AI Gateway or TypeSafe AI for the classifier, or any chat model through OpenRouter, OpenAI, Anthropic or your own OpenAI-compatible endpoint. • Your key is stored locally in your browser and never synced. • Posts go straight from your browser to the provider you chose. Typical cost with the default classifier is under one cent per thousand posts. MORE CONTROL • Per-author rules: never filter someone, or always hide them, straight from a post's ⋯ menu. • Blocked words, matched before any AI is asked, so they cost nothing. • Teach it: mark a wrong call and your correction is used as an example from then on. • Images, when they matter: photos can be described in words for the classifier, only for posts where the text leaves it in doubt. • Comments: replies are left alone by default. Turn on comment filtering if you want them judged too. • Bookmarks, notifications and messages are never filtered. • Optional "Teach X too": Sharp can send X's own "Not interested" for posts it hides on your Home timeline, so X's ranking learns as well. This is off by default. It is the one feature that acts as you on X, and the privacy policy describes exactly what it does. • Debug mode shows, for any post, exactly what was sent, what came back and how long it took. • Works at phone widths too, for Chromium browsers on Android that support extensions. YOUTUBE, THE QUIET WAY No AI involved. Hide Shorts, hide comments, and show, blur or hide thumbnails, on youtube.com and m.youtube.com. Greyscale is available for both X and YouTube. PERMISSIONS, AND WHY • x.com and twitter.com: to read the posts on the page and hide the ones that match. • youtube.com: to apply the page rules you switch on. Nothing on YouTube is read or sent anywhere. • Your AI provider's address: to send posts for a verdict. Providers other than the built-in ones ask for permission when you save them. • Storage: your settings, key and verdict cache, kept on your device. Sharp is free and open source (MIT). Source, issues and privacy policy: github.com/tshmieldev/sharp