Published June 16, 2026
Six months ago I started building a personal memory system because every note tool I’d tried had the same flaw: they assume you’ll happily feed them. Notion, Obsidian, Roam — they all begin with “you write, you tag, you link.” If you don’t curate, nothing grows. And curation is where most systems quietly die.
So I inverted the question: what if the memory curated itself? A local autonomous curator that digests the facts I drop during the day, folds them into the notes, removes duplicates, flags contradictions, archives the old. That was the last two months of work — the Librarian. It runs around the clock, every few minutes. Last week it crossed well past a thousand autonomous cycles, zero errors, several hundred facts integrated.
But the curator’s appetite was always bigger than my willingness to feed it. Every fact I wanted remembered, I still had to type. The whole point was to remove the friction between “I learned something” and “the memory has it.” And the friction had simply moved to the input.
Step 4 of the roadmap, finished today, was senses. Not in the sci-fi sense — in the literal one: passive capture of what’s already going through me on a normal day. Three modes, one rule.
First, voice. A keyboard shortcut: I hold it down, speak, release. A local speech model transcribes in a second or two and pastes wherever my cursor is. No assistant, no floating window, no “summon the AI.” It’s an input device with no opinion about what I just said.
Second, typing. A keyboard hook holds what I type in the active window in memory — nothing on disk, nothing leaving the process. When, and only when, I press Enter, that buffer goes to a local model that decides whether there’s anything worth remembering. Most of the time the answer is no: I just sent a message, ran a search, typed a command. When it’s yes, the fact enters the curator’s pipeline like any other source. The rule is the commit — I only learn what you confirm with Enter. Delete before Enter, nothing is kept. Switch windows without Enter, the buffer dies with the focus. A small choice that does a lot of work for privacy.
Third, screen. When I dwell on a window for more than ten seconds, a small process asks the operating system — through the same accessibility API that screen readers for blind users consume — what text is on screen. Not the pixels; that would be OCR, fragile and faintly creepy. What the app itself reports it is showing. A local model reads that text and tries to pull a single substantive topic, biased hard toward “nothing here.” If there’s something real, the memory notes that I was exposed to it today.
Here’s the part I want to land: a system that listens, reads, and watches cannot be passive about privacy. So the most important new feature isn’t any of the three senses. It’s the off switch.
There’s a small icon in the system tray with three modes. Low is the default — the memory learns from everything I type, dictate, and dwell on. Medium suspends voice and screen; only what I type with intent gets through. High suspends everything. One click. No restart, no menus, no “are you sure?” The listening just stops.
And underneath all three modes there’s a floor that never moves: password managers, banking domains, sign-in pages, password fields, and incognito windows are dropped before the sensors even fire — in every mode, and not tunable.
Today is day one of a seven-day pilot. A week of real use, then I review whether the system learned things I’d otherwise have forgotten without noticing. If it did, Step 4 ships. If not, I tune the prompts and start over.
What I notice already, two hours in: I haven’t typed a single fact into the memory by hand today. Not once. It has been growing in the background while I worked. The bottleneck moved upstream — from “I’d remember to type this” to “the system might learn this if I sit with it long enough.” Which is exactly where it belongs.
The next post is about that third sense in particular — why I turned down the obvious option, computer vision, for a much stranger one. That’s a design story worth its own page.
— Javier
EIDARA v2 is free. SUPER DARA is what comes next.