Okay, before I get into what all the cards and scores and archetypes actually mean, I want to explain the boring part first. Not because it’s boring to me — because if I were you, I’d want to know exactly what happens to my messages before I hand them over to some app made by a guy named Raif. So let’s do this like an actual manual, just… a manual I don’t hate reading.
What you actually do
You open WhatsApp, pick a chat, and export it — without media, so it’s just a text file. Then you share that file straight to Unrewind. That’s it. No login required to see your first result, no connecting your account, no giving anyone access to your actual WhatsApp. You’re moving one file, once, on purpose.
What happens before it ever leaves your phone
This is the part I actually care about most, so I’m not going to bury it.
Before that file goes anywhere, your phone replaces the names of the two people in the conversation — you and whoever you’re texting — with anonymous aliases. Person_A, Person_B, something like that. That’s the version that actually gets analyzed: the AI reading your messages never sees either of your real names, full stop. There’s one honest caveat to that claim, and one limitation we chose not to fake — I put both in the FAQ below instead of hiding them in fine print.
At the same time, anything that looks like a card number, an IBAN, a phone number, or a verification code gets stripped out. On your device. Before upload. Not after.
What happens on our end
Your message text gets read once. Just once. We use it to build a digest — a monthly summary plus a small set of verified quotes (never more than about 12 a month, or every message if the month was already quiet) — and after that digest exists, the raw message text is gone from our servers. Deleted. Not “hidden,” not “archived somewhere,” gone.
From then on, every card you see — Relationships, Profile, Tea — is built by reading that digest, not your original messages. That’s also why the second import of the same chat is so much faster and cheaper than the first: there’s nothing left to re-read, just a digest to update.
What you get back
Three things, and I’ll do a proper deep-dive on each one separately, but here’s the one-line version of each so this post isn’t a novel:
- Relationships reads one specific person and hands you the receipts — who’s actually putting in effort, what the pattern looks like, whether it’s warming up or cooling off.
- Profile looks across everyone you’ve imported and tells you what your patterns are. Are you the one who reaches out first? Do you go deep or stay surface-level? Turns out you have a type, and it’s not who you’re texting — it’s how you text.
- Tea (you’ll see it called Rewind on the website, same thing) is the yearly wrap-up. Your peak texting hour, your ghost score, the month that hit different. Built to screenshot and send to a friend.
What happens when something actually goes wrong
I’m not going to pretend everything works perfectly on the first try. Here’s the honest version of what happens when it doesn’t.
Your chat is huge and it’s taking a while. If you’re importing five years of texts with someone, we don’t try to swallow it all in one pass. It gets processed month by month, one pass per month, so a five-year chat means more work, not more risk. If one single month’s pass fails, it just gets retried on the next round — the whole import doesn’t fall over because September 2023 had a bad moment. You might be looking at a progress bar longer than you’d like. That’s not broken. That’s just a lot of texting.
The analysis comes back thin, or basically empty. This used to be a real bug in this app, and it actually happened to my own account once, so I’ll give it to you straight instead of pretending it never happens: the AI can occasionally choke on a huge, messy conversation and hand back almost nothing. Early on, the app would quietly accept that and call the job “done” anyway — which is exactly as bad as it sounds. That’s fixed now. If a result comes back incomplete, it gets retried automatically a few times, and if it’s still not right, the card is marked as needing another pass instead of lying to you about being finished. Hit “Analyze it” again and it actually does something this time.
You import the same chat twice. On purpose or by accident, doesn’t matter. Nothing bad happens — no duplicate person, no wasted quota. The app just recognizes the conversation and tells you you’ve already got this one.
Something’s off with the file itself. If what you shared isn’t actually a WhatsApp export — wrong file, corrupted export, that sort of thing — you get told that directly instead of watching a spinner do nothing forever. Re-export the chat (without media) and try again.
You lose your phone, or reinstall. Your on-device alias mapping resets, but the encrypted name backup I mentioned above means the analysis you already paid for doesn’t get stranded behind a meaningless label. Nothing needs to be redone from scratch.
None of this is “trust me, it just works.” It’s “here’s specifically what happens when it doesn’t, and here’s what we actually do about it” — which, if you’ve read this far, is kind of the whole point of this post.
Quick questions, straight answers
Since this is supposed to be the documentation-flavored part, here’s the stuff people actually ask, stated plainly.
Do you sell my data?
No. Never have, never will.
Do you use my messages to train an AI model?
No — it’s contractually blocked with every AI provider we use.
Do you know who my contacts are?
The AI never does — it only ever sees an alias. We do keep an encrypted backup of the real name behind that alias on our servers, purely so a reinstall doesn’t strand your paid analysis behind a meaningless label. It’s never sent to the model that reads your messages.
Do I need an account to try it?
No. You can see your first result before signing up, and decide after.
What happens if a third person is mentioned by name inside the chat?
That one we don’t catch. Telling a real name apart from an ordinary word in the middle of a sentence turns out to be genuinely hard, and a wrong guess would quietly corrupt your conversation — so we chose not to fake it.
What happens if my import gets stuck or the results look empty?
The app retries automatically a few times. If it still can’t produce a complete result, the card gets marked as needing another pass instead of a fake “done” — tap Analyze it again and it will actually reprocess, at no extra cost to you.
What if I import the same chat twice by accident?
Nothing bad happens. Unrewind recognizes it, tells you it’s already imported, and doesn’t touch your quota or create a duplicate person.
Why I built it this way
I’ve talked about this more in the manifesto, but the short version is: I went through a stretch where I was desperate enough to paste my own private conversations into a random chat window just to get an honest read. If I was going to build the tool I wish I’d had back then, it had to actually earn the trust it was asking for. Not “trust us” in a settings page nobody reads — trust that’s backed by the messages themselves disappearing once they’ve done their job.
Next up: I go through Relationships properly — every badge, every section, what it actually measures. For now, if you’ve got a chat sitting in your export folder — go see what it says.