An architecture drawn on a tissue paper

TRACE

We are building an actual brain — from the senses up.

Not an app. Not a recorder. Not a chatbot with search.
An organism that perceives your world, thinks about it while you sleep, and knows you better every single day.

founder architecture · 2026→ to advance

The premise

Every mind ever made runs one loop.

WORLD

⟶

EYES · EARSsight & sound carry almost everything

⟶

BRAIN

“Because we can’t manufacture other senses — and most of the information is obtained through these.” — the napkin

So we don’t invent new sensors. Video is the eye. Audio is the ear. The machine perceives the same world you do — and the entire problem moves to what a brain does with what it perceives.

Execution

This is not a pipeline. It is an anatomy.

WORLD

⟶

Video · Audioeyes & ears

⟶

black box 1EXTRACTIONa macro picture of what’s happening

⟶

coordinates
+ timestamps

⟶

BINDERphysicality × time → raw material

black box 2 · the breathing cellNODE CREATIONalways running · asks questions · 1+1=4

⟶

living, encrypted
nodes

⟶

black box 3INFORMATION SUPPLIERjust enough context

⟶

LLMdoes what an LLM does — the mouth

Three black boxes. Everything else is plumbing.
The black boxes are where the company lives.

Black box № 1 · Extraction

Perception is subtraction.

Discard the blurry. A brain doesn’t store its retina. Frames that carry nothing die at the gate.
Decipher signal from noise. Out of the torrent, a macro picture of what is actually happening — with its surrounding context attached.
Stamp everything with coordinates + timestamps. Specialized eyes — one helper per kind of task — write what they see, marked by where and when they saw it.

“Blurry frames discarder · context signal–noise decipherer → a macro picture of what’s happening.” — the napkin

Memory here is not addressed by content. It is addressed by spacetime. Where-and-when is the primary key — meaning attaches later, and can always find its way back.

The Binder

Many eyes. One account of the scene.

Every helper returns text marked by coordinates — a reading eye, a scene eye, a listening ear, each blind to the others.
The Binder fuses them physically and temporally: same place, same moment → same event. Scattered fragments become one continuous account of your world.
What comes out is not a node yet. It is raw material — “a first thought to marinate on.”

“Binding the coordinates — physicality & temporal — together. Note: this is not a complete Node. It’s raw material.” — the napkin

And because every thought carries its birth coordinates, every thought can be audited forever — walked back to the raw moment that produced it. Provenance isn’t a feature here. It’s the address system.

Raw material

The raw is sacred. That breaks biology’s ceiling.

Marination can fail. When it does, the failed thought is discarded — never the raw material underneath it.
Which means every thought is retryable. Better helpers, better questions, better models — point them at the same raw past and marinate again.
A human brain gets one consolidation pass, then its raw is gone forever — which is why memory confabulates and can never be re-examined.

“The marinated content that’s not useful is discarded — not the raw material.” — the napkin

So the system’s understanding of your past improves retroactively — your memory of last July gets sharper every year you own it. No biological brain, and no product on earth, does that. Time works for this store instead of against it.

1 + 1 = 4 ≠ 2

The crow and the stones.

Store three nodes — water, pot, stones — and all you ever get back is dead meta: water in pot, stones next to pot. Every database on earth stops here.

“What happens when a stone drops into the water?”

The water rises. — and “rises” existed in no node. No sensor recorded it. The question created data that storage never contained.

Intelligence is not storage. Intelligence is interrogation. That is what 1+1=4 means — and it is the entire reason this system exists.

Black box № 2 · The breathing cell

The crow asks because it is thirsty.

A brain with no need asks no questions. Remove the thirst and the crow stands beside the pitcher forever — nodes intact, deriving nothing.
The cell has one standing need: predict its human’s tomorrow.
Every surprise — everything today that its picture of you failed to anticipate — is water it couldn’t reach. Surprises become tonight’s question queue, ranked by how wrong they proved it.
So the night is never a swamp of everything. It is appetite with a ranking — the cell interrogates exactly where its model of you cracked, and nowhere else.

“A physical, living, breathing cell — always running in the background — can combine with fellow nodes to derive new things.” — the napkin

Always running. Always slightly wrong about you. Closing that gap while you sleep is what it does for a living.

The self-evaluating machine

Every new thought pays rent — or dies.

The crow validates “rises” by dropping the stone and watching. Our cell has something better than the world: the archive it swore never to delete.
A freshly marinated thought must predict a piece of the raw record it was not derived from. Derived from Monday? Then it must prove true in Thursday’s raw — moments it never touched.
Predicts nothing → discarded. The napkin’s discarder, given teeth: it doesn’t just remove the useless — it removes the false.
Nothing ungrounded is ever kept, so thoughts feeding on thoughts can’t drift into confident nonsense. The swamp is structurally impossible.

“The nightly must not be a swamp of information that biases more information down the drain — it must be a self-fulfilling, self-evaluating machine.” — the founder

The night doesn’t accumulate. It selects. Whatever survives to morning has already been tested against your real past.

Black box № 3 · Information Supplier

The LLM is just the mouth.

When you ask, the supplier walks the living nodes and hands the model just enough context — a small, local packet. Not search results. Not a dump. The already-thought thought.
Then, in the napkin’s own deliberately bored words: “LLM does what LLM does.” The hard thinking happened nights ago, in the cell.
Everything stays on the device. Nodes are encrypted — only the person using the app can decipher them. Your brain is yours.

“Information Supplier → just enough context → LLM does what LLM does. Local.” — the napkin

This is why the architecture is model-proof: every model is swappable; the organism is not. Each better LLM that ships makes TRACE more valuable — the intelligence lives in the memory, not the model.

Why you’re in this room

The black boxes are the company.

Solved problems: the senses, coordinate-stamped extraction, the binder, on-device encryption. Real engineering, already running on real hardware — but engineering.
Open problems: a cell that asks its own questions. Thoughts that grade themselves against a held-out past. “Just enough” context, found instead of searched. Nobody has built the organ that thinks unprompted and evaluates itself. Not the labs, not the giants.
And the giants can’t follow even once they see it: their business is the cloud, and this brain only works because it never leaves you.

“The HOW is the million-dollar question to
BUILD AN ACTUAL BRAIN.” — the founder

That HOW is the work. That work is the invitation.

WORLD⟶ EYES · EARS⟶ BRAIN

The first two boxes exist.
We are building the third.

A living memory that perceives your world, questions it while you sleep, discards its own bad thoughts, and speaks only when spoken to — private, local, and more yours every day.

T R A C E