On December 19, 2025, Rewind stopped remembering.
Rewind was the Mac app that recorded everything you saw and heard — locally, privately, searchably. People trusted it with years of their working lives. Then Meta acquired its parent company, and two weeks later the capture was switched off remotely. Users in the EU, UK, Brazil, and four other countries lost the service entirely, with about two weeks to export before deletion. Twenty months earlier, when the company pivoted to a hardware pendant, its founder had said there were "no plans to shut down" the app.
If that stings to read, here's the part that should sting more: Rewind wasn't an outlier. It was the rule.
The ledger
A short history of tools that promised to remember for you, and what happened to the people who believed them:
Atlas Recall (2016) raised $20.7M to build "a searchable photographic memory for your entire digital life." It shut down eleven months later. All user data deleted within nine days of the announcement.
Skiff sold end-to-end encrypted email and docs — privacy as the product. Notion acquired it in February 2024 and closed registrations the same day; six months later the service was gone. No reason for killing it was ever stated.
Omnivore, the open-source read-later app people used as a knowledge base, was acqui-hired in November 2024. Users got fourteen days to export before deletion.
Humane's AI Pin — $699 plus a subscription — was bricked with ten days' notice when HP bought the patents and the team in February 2025. Server data deleted. The devices still light up; they just no longer do anything.
Evernote — the original "remember everything" — took the other exit: acquired, staff cut to a skeleton crew overseas, renewal prices raised as much as 80%, and the free tier cut to fifty notes — with twenty years of your notes on the other side of the paywall. The product survived. As a balance sheet.
Khoj, the open-source "AI second brain," shut its cloud service in April 2026 with the most honest closing statement in the category: hand-building data integrations was "a long tail of effort," and there were "challenges in defensibility with all the big labs coming out with similar products."
That's one decade, one category. I'm leaving out Mailbox, Sunrise, Astro, Google's own Inbox, Heyday, Dot, and a dozen more. Across fifteen years of these products, the pattern holds with almost no exceptions: no consumer tool that promised to remember for you has ever survived as an independent business. They get acquired and killed, they pivot away from you, or they quietly become something that monetizes you.
Why this keeps happening
It would be comforting to blame bad founders. But these were capable teams with tens or hundreds of millions of dollars, and the same three forces got nearly all of them.
Remembering, by itself, doesn't retain. Capture-everything tools demo brilliantly and then fade from daily life, because their value shows up rarely and invisibly — you don't feel the search you didn't need to do. Heyday's founder, after winding down his memory assistant, put it better than any analyst: with general-purpose products like that, "either you somehow figure it out and you are ChatGPT, or you're just a magic toy." Atlas failed before modern AI existed; Rewind failed with modern AI built in. The technology wasn't the problem. The habit was.
The venture clock. Rewind raised at a $350M valuation on $707K of annual revenue — roughly 495 times sales. A number like that isn't a plan to serve you for twenty years; it's a promise, made to investors, that something dramatic will happen. When the dramatic thing arrives — a pivot, an acquisition, a wind-down — the users holding years of their memory in the product are a line item in someone else's transaction. The pattern is so consistent it has a tempo: from acquisition announcement to product death was sixteen days for Astro, three weeks for Humane's servers, six months for Skiff.
Free platforms eat the middle. The founder of Newton Mail, in the category's most honest post-mortem, wrote that a premium product simply couldn't "overcome the bundling & platform default advantages enjoyed by the large tech companies." That was about email clients in 2018. It's truer for memory in 2026, when every AI lab is bolting recall onto a free chatbot. Anything a platform can copy into its bundle eventually stops being a business — unless it's built somewhere the platforms structurally can't follow.
The cost always lands on the same person
Notice who paid the exit costs in every story above. Not the investors — Rewind's and Humane's backers were largely made whole or better. Not the founders, who mostly landed inside the acquirers. The cost landed on the person who did nothing wrong except believe the product's promise: the researcher with three years of reading in Omnivore, the family with their life recorded in a pendant, the writer with two decades in Evernote.
And the cost isn't just inconvenience. A to-do app dying loses you some checkboxes. A memory dying takes something closer to a part of you — the decision from last March, the promise from Tuesday, the thread that explains why. The more a tool knows about you, the more its shutdown costs you — and this category's history says shutdown is the default outcome.
The questions to ask before you trust the next one
Every few months a new product will offer to become your memory. Most will be genuinely impressive. Before you move your working life into one — including ours — ask four questions:
1. If the company's servers vanished tonight, would the app still work tomorrow? Not "is there an export button" — would it work. Search, browse, everything. If the answer is no, you're renting your memory.
2. Can you read your data without anyone's permission? Open formats on your own disk — files you can open in other software — or a proprietary blob on someone's cloud. One of these survives an acquisition. The other is Skiff.
3. What does the business need to happen for the product to keep existing? A tool priced to cover its costs can serve you indefinitely. A tool priced at 495x revenue needs an event. You are not the beneficiary of the event.
4. What, exactly, happens on acquisition day? Nobody will answer this one out loud. So don't accept an answer — demand an architecture that makes the question boring. Rewind's users had a promise: "no plans to shut down." Promises got kill-switched. Architectures can't be.
The uncomfortable summary of a decade of failures is that trust in this category cannot be a policy. It has to be a property of where the data physically lives. If your memory is on your machine, in formats you can read, in an app that runs without phoning home, then the company behind it can pivot, sell, or die — and you lose future updates, not your past.
That bar is high. Almost nothing on the market clears it. It's worth demanding anyway, because you're not choosing an app — you're choosing what happens to ten years of your working life on the day the app's story ends. And the one thing this graveyard proves is that the story always ends.
This is the bar we built Third Brain to clear: your memory lives in one SQLite file and open formats on your own machine, the AI runs locally by default, and the app works offline, forever, with export always available. We've raised nothing and answer to no investors, so there is no venture clock ticking behind it. And if we vanish tomorrow anyway, you keep everything — that's not a promise, it's the architecture.