Takt

BUILT WITH / NEURAL ENGINETensorFlow.js ↗

HIDDEN NEURONS: 24

TRAINABLE PARAMETERS: 412

RUNTIME: initializing_

loading local library_

Artificial neural network.
Runs on your device.

[ inspect engine source ↗ ]

Booting a small artificial nervous system. No account. No API key.

01 / THE CHAMBER

culture_001 / Observer

INITIALIZING

SEEK / OPEN FIELDTICK 000000

             initializing world_

@ organism+ energy# wall· trace22 × 12 cells

ENERGY FOUND: 000

COLLISIONS: 000

TRAINING STEPS: 0000

LAST REWARD: —

EXPLORATION: —

02 / INSIDE THE REFLEX

sensation → association → action

WEIGHTS / RANDOM

waiting for activations_

12 INPUTS → 24 HIDDEN UNITS → 4 Q-VALUES

ACTUAL MODEL ACTIVATIONS / NOT A SPIKE RECORDINGloss — / updates 0

03 / THE SIGNAL REGISTRY

Tokens outside. Reflexes within.

The largest pump.fun ecosystem tokens by market cap. Open a token. Initialize its own local culture. Train a reflex.

Reading the public registry…CoinGecko / pump.fun ecosystem ↗

TRACKED TOKENS: —

COMBINED MARKET CAP / USD: —

24H VOLUME / USD: —

DATA OBSERVED / UTC: —

loading public tokens_

Token identities and market data come from CoinGecko. Neural cultures are independent, browser-local experiments. Animated signal portraits are visual studies; they do not indicate biological activity, token ownership or project participation.

03 / THE CULTURE REGISTRY

Same world. Different beginnings.

6 LOCAL CULTURES

independent seeds / independent weights

Every culture starts untrained. Portraits identify seeds; performance comes from your experiments. Nothing is shared between visitors.

04 / THE EXPERIMENT RECORD

What changed?

LAST 20 / SAVED LOCALLY

CULTUREEXPERIMENTRESULTTIME

No experiments recorded. Train a culture, then test what it learned.

05 / FIELD NOTES

A reflex is a memory of consequence.

[ READ ME ]

001 / THE PREMISE

Nothing knows where to go.

A randomly initialized network inhabits a tiny world. It sees energy, nearby barriers and its surroundings. Four predicted action values become one movement. An outcome becomes a training example.

002 / THE MECHANISM

Keep the consequence.

Energy is rewarded. Collisions are penalized. A distance-based reward guides progress. The sandbox explores four possible actions per state (Dyna). Replay revisits those transitions; gradient updates change the network. Testing freezes it and disables exploration.

003 / THE BOUNDARY

Artificial. Local. Inspectable.

TensorFlow.js runs a small neural network on your device. These are computational organisms, not living cells or conscious beings. No biological equipment, model API or token is connected.

TensorFlow.js documentation ↗

OBSERVATION ──→ ACTION ──→ CONSEQUENCE ──→ MEMORY ──→ WEIGHT UPDATE ──┐
↑___________________________________________________________│

Life begins as noise.