Martial Tipsey

Worked example · private memory

The memory my AI tools share.

Every chat with an AI starts from nothing. Close the window and the last two hours are gone. So I built one place to keep the memory, my open brain, and gave every tool I use the same abilities: save a memory, and find one again. It doesn't send anything, post anything, or delete anything. And that simplicity is the point.

The short version

  1. I don't file anything

    At the end of a session the assistant writes a short summary and saves it.

  2. It gets sorted and labelled

    Stored so it can be found later by what it meant, not what it was called.

  3. It stops there

    Two things only: put a memory in, and find one.

  4. The next chat starts informed

    A new chat, a different AI. It asks, and the right memories come back.

What actually happensEight steps, start to finish.

Any AI client all reach the same place Or a small program adds memories directly Sorted and tagged before it's stored Something I dictated, or a work memory? Technical, or everyday? Topic, a few labels, one-line summary Original text kept exactly as written Turned into a fingerprint of what the memory means A question finds its nearest memories closest first Memories in, memories out smart storage — experience in a box that's the simplicity Experience where it counts Instagram, Substack, Threads
  1. Every session ends the same way

    Not a recording of the whole conversation, a short summary. What we worked on, what got decided, what broke, what's still open. Written so a tool that knows nothing can pick it up cold.

  2. It doesn't matter which AI I was talking to

    They all reach the same place. Different apps, different companies, one memory. A small program can drop memories in too, which is how a backlog of older summaries got loaded without me retyping any of it.

  3. It sorts the memory before it saves it

    First question: is this something I dictated myself, or a memory from a working session? Mine take a different path and are marked as mine, so an assistant's summary never gets mistaken for something I said.

  4. Then it gets labelled

    The topic, a handful of labels, and a one-line summary, filed alongside the memory, never instead of it. The memory itself is kept exactly as written. The summary is the label on the box, not what's in it.

  5. The words get turned into a fingerprint

    This is the part that makes it useful. Each memory is turned into a kind of fingerprint of what it means, so I can search for the sense of a thing rather than the exact words. I can ask about "the thing that kept timing out" and it finds the right memory even if the memory never used that word.

  6. Finding it again works the same way backwards

    A question gets turned into a fingerprint too, and the closest memories come back, nearest first. Anything that isn't a genuine match is dropped rather than padded out — "nothing relevant" is an answer it's allowed to give.

  7. Memories in, memories out. That's the simplicity

    It's smart data storage and for AI, it's basically experience in a box.

  8. Experience where it counts

    The same memories feed my other work. My pipelines for Instagram, Substack, and Threads read them, and what I change afterwards goes back in. It's the experience the AI needs to work better and smarter. For me. Read more about my social pipeline here.

My memories are my own, and so is this

It runs on a private computer I rent and look after myself. The only things that can reach it are the handful of tools I've approved.

And it's mine in the way that actually counts: Nothing locked inside someone else's app or service, waiting on their pricing to change or their company to be sold. And it's downloadable at any time to wherever I want it.

What do other people see?

Nothing, unless you want them to. There's no public view, no shared link, nothing handed around. What's in my memories stays between me, my AI, and my open brain.

The same shape, in a business

Nothing above is really about AI, it just makes it easier to do. The pattern is: capture what was decided at the moment it's decided, smartly store it in one place in a form that can be quickly searched by relevancy, and let it be surfaced at the right moment so future actions and decisions are properly informed.

Every business already has a version of this problem. Everything you told the last person who did this job left with them. The reason you priced that customer the way you did lives in one person's head, and they're on holiday. A new hire takes three months to be useful because the answers exist but nobody can find them. Nobody is checking the knowledgebase. That's the same gap: the knowledge was created, and then it wasn't kept anywhere you could easily ask. It reads back what it remembers; it doesn't act on it — the same boundary as everything else I build.

Interested in something like this? If there's something you keep doing by hand, let me know and I'll tell you what it would take to automate — the work moves to the AI, the accountability stays with you. And if it isn't worth building, or if something off the shelf would do it better, I'll tell you that instead.

Tell me about it

Running a business and suspect there's more than one of these in your week? The diagnostic is built for exactly that.