Sharp by tshmieldev

Sharp cuts the slop out of your feeds. You describe what you want to see, or never see again, in plain language, and your own AI model does the judging.

X
• Filter in your own words: "hide engagement bait and crypto threads", or "only keep software engineering". Whatever you'd tell a person.
• Bring your own model. OpenRouter, OpenAI, Anthropic, or any OpenAI-compatible endpoint. Your key, your bill, your choice of model.
• Hidden posts collapse to a one-line banner with the author and the reason. Show any of them with one click. Or hide them completely when the model is sure; close calls always keep their banner.
• Teach it. Mark a hidden post as a wrong call, or pick "Hide posts like this" from a post's menu.
• Teach X too. Sharp can tell X you're not interested in what it hides, so X's own ranking learns alongside the model.
• Per-author rules and blocked words, applied before the model is asked, so they cost nothing.
• Thread escape hatch: one click shows every reply in a thread, filters off.
• Spending presets: choose how many posts go in each request and how many requests run at once.
• Activity: posts hidden, requests, tokens, and an estimate of time saved.

YouTube
• Hide Shorts everywhere: the shelf, the sidebar entry, and any Short in feeds, search or notifications.
• Hide comments.
• Thumbnails shown, blurred, or hidden. Hidden keeps the duration badge so you still know what you're clicking.

Both
• Greyscale the chrome, the content, or both. Less distractions, same information.

Private by construction
There is no Sharp account and no Sharp server. Your API key stays in your browser's extension storage. No data is collected by Sharp.

And it's open source: github.com/tshmieldev/sharp