Disclaimer: I am the co-founder of Mneme, so I am not neutral, and I will not pretend to be. But everything below is real: the problem, the research, and the night I describe. This is an honest look at what the product does well, written by the most biased person available.
Three weeks ago I sat down to do work that I had already done once. I am a founder, and I write code. My job is one part building, one part selling, and one part reading: market research, competitor analysis, regulatory PDFs, analyst reports, and a constant stream of GitHub repositories that could maybe save me a week one day. That night I was preparing a competitive analysis for a partner meeting. I had read everything I needed a few weeks before. I had saved everything I needed. So I opened Claude and started typing the context from memory, and twenty minutes later I was still typing.
The analyst report was somewhere. In my bookmarks? In Notion? In an email I sent to myself with the subject "read this"? I had already explained half of this market to ChatGPT for a different document, but that conversation was lost in the history. And the repository that did exactly the data collection I needed, I remembered starring it. I did not remember its name.
I had done this research once, and now I was doing it again. That night I was angry enough to sit down and count.
The graveyard
The count: 752 saved articles. 140 links in a folder called "read later", which is a lie. 18 Notion pages with names like "Competitor notes v2 FINAL". 148 starred repositories, starred for reasons I don't remember anymore.
If you live in the second brain world (Notion, Obsidian, Roam before that) you know this shape. The promise was compounding knowledge: capture everything, and your notes will start working for you. The capture part worked very well, because capturing is easy, feels productive, and takes two seconds. The retrieval part never happened.
Here is the uncomfortable truth I had to accept: filing is not remembering. A note you never open again is not a memory, it is a burial. Most second brains are write-only memory, a graveyard with good metadata. And the problem is not discipline. The second brain did not fail because we saved too little or tagged too lazily. It failed because it became one more place to look, one more app fighting for our attention. (My co-founder and I had a real fight about this. He created the 13th page in our Notion, and I told him that a 13th page is not memory anymore, because now I also need to remember what the other 12 pages have. He said the page was necessary. We were both right, and that is exactly the problem.)
For years I made peace with this. Then the place where I actually work moved, and everything got worse.
The amnesia
I don't start my work in an empty document anymore. I start in a chat box. Claude for strategy and writing, Cursor inside the codebase, ChatGPT for quick questions, Gemini when the context is very long. If you are reading this post, your setup is probably a version of mine, and if it is not yet, it will be soon. And every one of these tools starts every session knowing nothing about me.
Let me give a real example. After every meeting, I write the follow-up emails with Claude, because Claude is surprisingly good at empathy and tone. So over time, Claude learned my meetings, my clients, my whole email history. But none of that was in Notion. And none of it was in ChatGPT, so when I asked ChatGPT to score my sales leads, it scored them like a stranger. For ChatGPT, they really were strangers.
So I became the wire between my own tools. Copy from here, paste there. Upload the same PDF again. Explain the same positioning again, in slightly worse words than the first time. I was keeping a mental map of which model knows what, and the map was wrong half the time, because I am a human running a company and doing nine things at once. I was working as a router between my own AI tools, and the router was dropping packets.
You have the knowledge, the model has the reasoning, and there is no wire between them.
By the way, this is not my special problem. It is a timing problem. AI moved faster than our tools for keeping context, and everyone who works seriously with these models hits the same wall. Some people simply hit it later. A smarter model with no memory is only a smarter stranger.
The missing layer
There is one more person in this story: my co-founder, who slowly became my search engine. I ask him things like, what was the name of that company we met in March? Who was the founder of that product? When did we share the LinkedIn post about that VC investment? He usually knows. But he is also a human, also tired, also saving things into his own 13 Notion pages. Many times I thought the same thing: if Claude or ChatGPT had this data, I would have the answer in two seconds, and nobody's day would be interrupted.
That thought changed how I saw the whole problem. This was never a note-taking problem. For twenty years we built better and better boxes for knowledge: folders, wikis, graphs, vaults. But a box is a box. It holds what you give it, and it cannot answer with it. Now, for the first time, we have software that can actually reason, and it has nothing of yours to reason over.
Think about how your own memory works. You don't search your brain with keywords. Relevant things simply come up. You read a sentence, and something you learned three years ago appears by itself. Memory that only answers to search is not memory, it is storage.
So what would the missing piece look like? Not another app to check. A layer. Neutral, feeding every model you use, owned by none of them. Always on, so that remembering does not depend on you remembering to remember. And, most important, something your agents can query directly, the same way they already query the web. Somewhere in this thinking, I decided I needed this layer more than I needed another note app.
One thing, then I stop
First, I looked for the layer, because building something new is expensive and I am lazy in the good engineering way. I tried the memory features inside the chat tools. They remember your chats, which is nice, but they know nothing about your browser, your PDFs, or your saved research, and whatever one tool learns stays locked inside that tool. I tried the AI note apps too. They made my box smarter, but it was still a box, and my agents could not reach into it. After a few weeks it was clear that the layer I wanted did not exist. So we started building it ourselves. It is called Mneme, and yes, this is the part where I remind you that I am biased.
Everything dumped in one place; the topics on the left built themselves. No 13th page.
I will show you exactly one thing it did for me, and then I will stop on purpose. A few weeks ago we had a difficult meeting coming with a possible vendor. The open problem sat between our legal, compliance, and brand teams: questions about the EU AI Act, personal data rules, an open trademark issue, and a long email thread between our lawyer and their legal team. For two weeks I had been saving everything about it into Mneme without thinking much: regulation articles, risk notes, the email thread, an AI Act summary I found on some random blog, even my own half-finished thoughts after calls.
The night before the meeting, it was around 11pm and I was, honestly, hopeless. Normally this is the moment where I open ten tabs and start searching through everything again. Instead, I pressed ⌘K and typed: "prepare my notes for tomorrow's vendor meeting: their legal team's concerns, our compliance position, and the open trademark question." What came back was a meeting prep document built only from what I had saved. Every open issue was listed, with what their lawyers wrote and in which email, next to the exact AI Act articles that mattered, all cited back to the sources and ready to insert into Claude. I read it twice, fixed two small things, and went to sleep. The next morning I was the most prepared person in that room, and the preparation took minutes instead of a night.
For the builders reading this: your memory is exposed as a personal MCP endpoint. You connect Claude, Cursor, or Gemini once, and every tool that speaks the protocol can recall what you saved. With your permission, revocable any time, never used for training.
Mneme does other things too. There is a version of this trick for podcasts, and one for coding agents that I really like, and I will not show them here. This post is about the problem. The rest you should see yourself.
Mneme is in early access → yourmneme.com
Everyone is going to need one
Here is what I think will happen, and you can hold me accountable for it. Very soon, every knowledge worker, builder, and content creator will need a memory layer between their browsing, their Notion, their phone, and their AIs. Not because it is a nice productivity trick, but because agents are moving from answering questions to doing real work, and real work runs on context. An agent without your context is a very fast intern on their first day, every day.
Nobody typed this. The layer wrote it from a few weeks of normal work — and updated itself 32 minutes ago.
The big AI companies will ship their own memory, locked inside their own products. The note apps will put AI on top of their boxes. There will be many options, and honestly, you should start with one of them early, whichever you choose, because memory compounds: every month you wait is a month of yourself that no tool will ever get back. But I believe the winning layer will be the neutral one, the one that belongs to you and speaks to every tool you use. That is the bet behind Mneme, and I am completely biased about it. Maybe it will be the best one among them, but that part is not mine to say. Request access, feed it one week of your real work, and decide for yourself: yourmneme.com.
The name, if you are curious, comes from the Greek muse of memory. Mneme: the persistent effect of past experience. That is the whole pitch, actually. Your experience should persist.
I'm Furkan, the founder of Mneme. I will be writing about memory, AI agents, and the honest parts of building a product while also running a company. If you read this far, the most useful thing you can do is try it and tell me what feels wrong (I read every message).
Follow the build on X · request early access at yourmneme.com
Top comments (0)