rewind vs innernet
rewind was the best consumer memory product anyone had built. it isn't a competitor any more. it's the argument.
the short version
Rewind is the most important comparison on this page, and it isn't a competitor any more. It's the argument.
Rewind built the best consumer memory product anyone had built: perfect recall of everything you saw and heard on a Mac, then Ask Rewind on top of it. It became Limitless, Limitless was acquired by Meta, and on 19 December 2025 the Rewind app stopped working. The domain now hosts an unaffiliated AI-tools site with no connection to the original company.
So: the recall was excellent, the team was good, the funding was there, and the memory is gone. Every memory product should be evaluated against that ending, including ours.
what rewind was building
From 2020, a searchable recording of your digital life on macOS — screen, audio, meetings, everything, stored locally, with GPT-4 layered on in 2023 so you could ask it questions. Then the Pendant in late 2023, a wearable that recorded conversations. In 2024 the company rebranded to Limitless and moved toward cloud-based personal AI and the hardware, with the original Mac app entering maintenance mode. Meta acquired Limitless in 2025; the Rewind app ceased functioning in December of that year.
It was a real achievement and it's worth saying plainly: on raw recall, it was ahead of everything.
what innernet is building
A personal memory layer that lives outside every tool. npx innernet connects the AI tools on your machine to one memory. netti files what accumulates into Context Maps — dimensions, nodes, commits, branches. Every fact carries how far it may travel. Export everything or delete everything, on demand.
We are not trying to record your life. We're trying to hold your thinking, in a form other tools can read, in a way that doesn't depend on us being the only door to it.
what the ending actually teaches
1. memory has a survivorship requirement that other software doesn't
If a notes app shuts down, you lose a notes app. If a memory shuts down, you lose the record — and the longer it worked, the more you lose. That asymmetry should change how you choose. The first question about any memory product is not how good the recall is. It's what happens to the memory when the company changes direction.
2. capture everything is the most fragile promise in the category
Total recall is expensive to run, heavy to store, and it makes the product the only thing that can read its own archive. That's a great moat right up until the moment it becomes a liability on someone's balance sheet. A memory that's smaller, structured and readable by tools you don't own is less impressive and much harder to strand.
3. the exit risk is a product design question, not a trust question
"We won't do that" isn't an answer. The answers that count are structural: can you take everything out, right now, without asking; can something other than the vendor's own app read it; is deletion real. innernet's are — export everything and delete everything on demand, deletion immediate with backups purged within 30 days, memory never sold, and an open MCP endpoint at innernet.live/api/mcp so the memory is reachable by tools we don't control.
side by side
| rewind (as it was) | innernet | |
|---|---|---|
| What it captured | everything on screen and audio | what accumulates in your AI work |
| Where it lived | on-device archive, then cloud | hosted, readable by any connected tool |
| Who could read it | the Rewind app | any MCP-capable tool, by URL |
| Shape of memory | a recording you search | dimensions, nodes and conclusions |
| Status | app ceased functioning, Dec 2025 | live |
| Exit path | — | export everything, delete everything, on demand |
where rewind was genuinely better
- Recall. Nothing since has matched a complete searchable record of everything you saw. That was a remarkable piece of engineering.
- Zero capture effort. You didn't have to do anything. Nothing we build will ever be that frictionless.
- On-device, originally. The early architecture kept the archive on your machine, which was the right instinct.
where innernet is different by design
- We hold conclusions, not footage. Smaller, structured, and meaningful without a search query.
- The memory is readable by tools we don't own. That's the property that makes it survivable.
- You can leave with everything at any moment. Not a promise about our intentions — a feature.
- Every fact carries how far it may travel. Recording everything and deciding disclosure per fact are opposite philosophies about what memory is for.
which one you want
Rewind isn't available, so this isn't a choice. But if what you loved about it was total on-device recall, the honest answer is that innernet isn't that and doesn't intend to be — look at the screen-recording tools that picked up the thread.
Use innernet if what you actually wanted from Rewind was for your tools to stop making you re-explain your own work, and you'd like that to survive the next acquisition.
the question we actually get
"How do I know innernet won't just be the next Rewind?"
You don't, and anyone who answers that question with reassurance is selling. What you can check instead: whether you can export everything today, whether deletion is real, and whether anything other than our own app can read your memory. Ours are yes, yes, and yes — an open MCP endpoint means your memory is reachable by tools we don't control, which is the only version of that promise that survives us. Ask every memory company those three questions and watch which ones change the subject.
Accurate as of 3 September 2026. Timeline drawn from Rewind's own "what happened to Rewind" page and public reporting on the Limitless acquisition. The rewind.ai domain now hosts an unaffiliated AI tools product. We have no quarrel with the original team — they built something excellent.
this is what innernet remembers. you can keep your own.
innernet is memory your ai tools read live — projects, decisions, the things you’d otherwise re-explain every session. we write these letters out of ours.