Softwarechrome extensionxocrtesseractopen source
Rewind for X, a free MIT-licensed Chrome extension, records the posts you scroll past on your X home timeline and searches the text inside their images with Tesseract.js
It counts a post only once more than half of it was on screen, keeps at most 3,000 entries in Chrome's local storage, and opens a clicked post in a new tab so the feed you were reading never reloads.
You see a post worth keeping. You scroll. You pull to refresh, or X does it for you, and the post is gone. The README opens with that exact pain, in Chinese: “X 的首页推荐一刷新，刚才看到的好帖子就找不回来了.” Roughly, the good post you just saw can’t be found once the feed refreshes.
The fix is a Chrome extension that watches what you scroll past. The README says it “quietly records” every post you’ve seen on the home feed, and the rule for “seen” is strict: a post counts only when more than half of it showed on screen. Scrolling by at full speed doesn’t fill your archive with junk (a refreshing change from a feed that does that on its own).
For each post it saves the author, text, images, quotes, reposters and links. Both the recommended and following feeds are covered. Ads, post detail pages and profile pages are not recorded, so this is a log of your timeline and nothing else.
Open the log from a button at the bottom right of the page, the toolbar icon, or Alt+Shift+R. Newest posts come first, grouped by browsing period. There is keyword search plus filters for images, videos, quotes and reposts, and the README says Chinese search works whether or not there are spaces around the words.
One small decision is the one I like most. Clicking a post in the log opens it in a new tab, so your home feed doesn’t reload and shuffle itself into something you can’t recognize. A history tool that wrecks your place in the thing it’s recording would be an odd product.
Storage is modest on purpose. The README says the log lives in chrome.storage.local, with “最多 3000 条，可一键清空,” which means at most 3,000 entries and a one-click clear. It also says “数据只在本机,” that the data stays on this machine. Post number 3,001 pushes out an old one, so this is a short memory, not an archive.
The image search is the clever part. Text inside attached images becomes searchable through Tesseract.js 7, which the README lists under Apache-2.0. It runs in an offscreen page, and background.js works through images one at a time. The README says recognition “在本机完成,” meaning it finishes on your machine. The catch is the first run: it downloads a few MB of recognition data from jsDelivr, so you need a connection once before “offline” is true.
Install is either the Chrome Web Store listing the README links to, or the developer route: open chrome://extensions, turn on Developer mode, choose “Load unpacked,” pick the repo folder, then open or refresh x.com/home. The repo is MIT-licensed and showed 415 stars when I read it.
Two warnings before you rely on it. The README is in Simplified Chinese, though it points to an English homepage at rewindforx.com. And it says plainly: “非 X 官方产品，与 X Corp. 无关,” not an official X product and unrelated to X Corp. X can change its page markup whenever it likes, and an unofficial extension has no say in that.
The README never says what happens to the log when X reshuffles its page markup, and the permissions the extension asks for aren’t listed there either.
Check the log before you trust it.