The review-and-passback loop for your AI drafts
Your AI hands back a long answer and 82% of it is great. The other 18% is five specific things wrong in five specific places. PassbackAI is where you mark each fix on the exact passage it belongs to and pass the whole batch back in one round-trip — so the model lands every correction on its target instead of rewriting the part you liked. Most of the time an answer misses because your follow-up wasn't precise enough about the model's last message. This is how you finally respond precisely to the text your AI gave you.
It works with whatever AI you already use
PassbackAI isn't tied to one model. Whatever gave you the draft — ChatGPT, Claude, Cursor, Gemini — you paste the answer, mark it up, and drop the marked-up batch back into the same chat you were already in. The loop is identical everywhere: one surface for the corrections, one round-trip back, every fix anchored to the span it's about.
Claude is simply the deepest-wired example. Because our MCP server is live in production today — not on a someday roadmap — a draft can route straight from Claude into PassbackAI, and your marked-up answer routes straight back, with no copy and no tab-switch. That's one convenient version of the loop, not the whole story. The point is that it's cross-platform, so no single vendor's chat box owns your feedback step.
Which answers the question everyone asks: "won't Claude or ChatGPT just build this in?" Maybe one of them ships a highlight tool eventually. It won't make this redundant, because the step where you correct a draft sits between every model and your next prompt — and you don't use only one model. PassbackAI is the surface your AI routes into, whichever AI it happens to be.
Where your document actually goes
Here's the honest version, path by path — because "where does our document go?" is the question that decides whether you can actually use a tool like this at work.
The tool you'll use by default — paste, mark up, export — runs entirely in your browser. Nothing you paste is uploaded; your document lives in this tab, on a shelf of your twenty most recent docs (plus anything you pin) — older drafts quietly make room. Share it with a plain link and the whole document rides inside the link itself, so there's still nothing sent to a server. You don't have to take my word for it: open DevTools → Network and watch as you paste and mark up — no request carries the doc.
There are exactly two ways a document leaves your browser, and both are deliberate moves you make, never defaults — and they land differently. Choose a private link so someone can respond, and your browser encrypts the document before it goes: the key stays in the link and never reaches us, so it's end-to-end encrypted and I genuinely can't read it. Route a document to a reviewer through your connected AI, and that copy is stored on our server under server-managed keys instead — encrypted, but not end-to-end, so I won't pretend we can't read a doc you routed. Naming that difference plainly is the whole point. Full posture at /privacy.
Who's behind it
Elad Diamant — one person, in Tel Aviv. I built PassbackAI after the fourth time I typed "not that one, the other one" into a chat box. Before this I spent a decade in hospitality SaaS, watching sharp reviewers — PMs, editors, legal — give up on points four and five of every doc review for the same reason every time: retyping each correction into a box was hard enough work that the last two never got written. The next reader was sometimes a person and sometimes a model; the missing tool was the same either way.
It's free to use. There's no account. I read every note that comes in — if something's broken, unclear, or missing, email me and you'll hear back from me, not a queue.
But don't take the pitch on faith. Paste a long answer your AI just gave you and mark it up — feel how fast five fixes go back in one shot. That thirty seconds tells you more than this page can. Start here →