PassbackAI

You spent forty-seven minutes typing "yes, but the third bullet under the second heading — and also the FAQ tone, and the part where you invented a citation" into Claude. Claude apologized, regenerated the whole thing, and made it worse in three new ways. You closed the tab. You are not alone. The cruel twist isn't that LLMs can't take feedback. They take feedback fine. The cruel twist is that you have to type the feedback as one long sprawling paragraph, in prose, in order to give it. Your editor brain wants to click-and-mark. The chat box wants you to write an essay about your essay. OMG. OMGfixMD is the part of the workflow that was missing. You copy the response, click "Review" on a small toast, and the text opens in omgfixmd.com — a browser-only review surface where you highlight specific passages and leave a one-line note on each. When you're done, you copy the cleaned-up follow-up prompt and paste it back into the chat. No more typing "the third bullet, and also the tone in paragraph two…" by hand. Now the part the privacy paragraph usually buries: Nothing leaves your browser. OMGfixMD has no backend. The extension has no backend. There is no database to breach because there is no database. Your copied text moves from the LLM tab → the extension's in-memory state → the omgfixmd.com tab, then is cleared. Everything happens inside your own machine, in RAM, never written to disk. No analytics, no telemetry, no third parties. The only thing the extension stores is one boolean per origin: "you clicked ✕ to silence me on this site, so I'll stay quiet here for the rest of this browser session." That's the entire data footprint. What works today: ✅ Capture on any LLM tool that copies long text via the standard clipboard API — Claude, ChatGPT, Gemini, Perplexity, plus whatever ships next month. ✅ Inline annotation in omgfixmd.com: highlight, label, comment, all client-side. ✅ A refined prompt waiting in your clipboard, ready for one ⌘V back into the chat. What doesn't work yet, said plainly: ❌ Auto-pasting the refined prompt back into the source chat. That's the next step. Until then, you do one paste. We are still saving you five.