Why I'm building Scribbles.

There's a bigger version of the AI memory problem that doesn't get talked about much.

The conversation usually starts here: AI is great at reasoning, but it forgets you between conversations, so we need to give it a memory. A whole category of products has shipped around that idea — ingesting your chats, files, notes, and other sources, extracting memories, and letting you search or chat across them. Each is solving a real problem.

But most of that work lands in the same place: a searchable pile. The memories are there, embedded and retrievable, but they're not things you can look at and make sense of. You can ask questions, and the system will surface chunks. You can't scroll your books. You can't sort your trips. You can't fix a wrong year on a record you care about. The output of all this ingestion is a better retrieval surface, not a coherent picture of you.

We think the more interesting shape is upstream of that. A place where everything lives in one structured, readable form. Not as chunks in a vector store, but as real records with real fields. We call it a personal context layer. A single, structured, human-readable place where the things that make you you live as clean records. Books you've read, with authors and years and your ratings. People you know, with what you last talked about and when. Trips you've taken, with dates and places. Decisions you've made, drafts you're working on, receipts you've filed. Yours to see, browse, sort, edit, and trust. And usable by anything that needs to know who you are, including any model working on your behalf.

The same shape exists at the team scale, in Slack threads and customer notes and internal wikis and ten people's heads. The answer there is the same kind of thing.

Getting here has meant pulling apart a few assumptions that most of the existing stack takes for granted. Capture has to be frictionless, but the output has to be structured. Memory has to be browseable, not just searchable. Schema has to come out of messy natural language without asking users to think in databases. We've been at it for a while, and a lot of it still feels like an open problem.

What Scribbles is trying to get right.

Steerable. You decide what the system remembers. A lot of AI memory works by quietly absorbing everything you say into a store you can't see. We wanted the opposite. Nothing meaningful enters your memory unless you want it to. The raw scribbles are still there if you ever need to go back and query them, but by default, your memory only holds the things that actually matter to you.

Human-readable. Memory should look the way memory feels. A list of the books you've read. A roster of the people you know. A timeline of the trips you've taken. Structured records with real fields, not graphs nobody can parse. You should be able to browse your own mind.

Self-organizing. Maintenance is what kills most personal knowledge systems. The structure is great for a week, then you fall off, and the whole thing collapses. So Scribbles has to do the extracting, the filing, and the linking on its own. You drop things in, and the system works out what they are and where they belong. The more you use it, the more organized it gets, not the other way around.

Rich experiences. (Still experimenting here.) When the activity is something you'd already want to remember — reading an article, going through a research report, watching a video — we let you do it inside Scribbles, so the highlights and notes are already part of your memory by the time you're done. Optional, not required. The capture works just as well if you keep doing all of that elsewhere and send things in afterward. But for the moments where the activity and the memory are the same thing, it feels powerful to have them in one place.

If this resonates.

Scribbles is early. A lot of the edges are still rough. If the idea speaks to you and you want to try it, I'd love that. And if you want to tell me what you think — good, bad, confusing, missing, broken — email me at aneeq@usescribbles.com or DM me on X at @aneequrrehman. I read everything, and I reply to most of it.