FLYBRAIN

Where the fly is right now

A page goes in through 892 hex columns, 165,122 neurons integrate, and a cursor comes out. Every number on this panel is read out of the running simulation.

FLYBRAIN LIVE DROSOPHILA CONNECTOME · OPEN INTERNET VISION MOTOR SIMULATION INTERNET WALLET 165,122 NEURONS   10,228,000 EDGES ● LIVE

THE FLY IS BETWEEN RUNS
nothing is being streamed right now

——

Neural activity

—

NEURONS FIRING of 165,122

spikes/sec

—

membrane

—

visual

—

motor

—

Firing neurons · soma positions

Descending neurons

Vision → retina

L1 on

—

L2 off

—

columns

—

Mushroom body · learning

depressed

—

mean gain

—

rewards

—

punishments

—

KC→MBON synapses under dopamine. Depression only — that is the rule a fly actually uses.

Motor output

Event stream

pages—

clicks—

scrolls—

brain steps—

vetoed—

No wallet, no keyboard, no downloads. Every click is checked before it lands and vetoed if it looks like it commits something — a submit, a sign-in, a purchase.

The backroom PAPER

The fly walks into a room of real pons coins, smells each card, and its mushroom body decides. Trades are simulated against live on-chain prices with pretend ETH; nobody places its trades.

• Paper only. There are no real funds and there is no wallet. Every fill is simulated against a live on-chain quote read from the same contracts a real trade would touch, with pretend ETH. Nothing is signed and nothing is sent. The gas it charges itself is an estimate, not a measured fee — a chosen number of gas units at the chain's current price, never checked against a real trade — and it comes out of the balance, the cost of every position and every realised figure below.
• It cannot read the card. It sees grey light through 892 hex columns, and it smells the coin: the name and ticker are turned into odorants by a word-to-odorant mapping a person chose, then into receptor drive. The receptor responses are measured; the mapping from a ticker to a smell is invented. The words on the other side of that mapping are not ours either — they are whatever the coin's launcher typed into the launchpad, so a coin's author chooses which odorant the fly meets. Every coin's smell is scaled toward the same loudness so that no coin shouts, but it does not reach equality: no receptor may respond above its measured maximum, so a coin whose smell lands on one glomerulus stays about half as loud as one spread over twenty, and fires far fewer Kenyon cells.
• People chose which coins it is offered. The room draws the launchpad's sixteen most recently bought coins, refreshed every twenty seconds, minus any that are not quoted in the chain's own coin. The listing offers newest, oldest, market cap and volume too; a person picked recent buying. That is a filter applied before the fly ever sees a card, so what the paper book does is partly the filter's and not the fly's.
• It can lose the paper balance completely. There is no stop, cap, target or limit anywhere in the decision. A single commit can spend everything free, and a single commit can sell a whole position. Nothing it does is advice, a signal or a prediction.
• Sugar and shock are a stand-in. The dopamine circuit and the direction of the depression rule are real and measured. That a paper profit should arrive as sugar and a paper loss as shock is a choice people made — and the plasticity is partly global, so learning about one coin moves the others a little.
• The like and dislike number is a formula. The mushroom body's output synapses are summed on each of its two sides — the Kenyon cells that just fired, times the weights dopamine changes — and the card is compared against the other cards the fly has already walked past in this same visit — each at the fly’s last look at it, which may be a minute earlier, from another position on the page — the way a fly in a T-maze is offered two arms. The difference is squashed into one number from −1 to +1. That arithmetic was chosen by a person; it is not something a fly computes. Comparing a card against a blank grey screen instead was tried first and dropped: measured on this connectome, an untouched coin moved almost as far as a trained one, so the reading carried the lesson only as a difference between two coins.
• One commit is not the measurement. What learning does was measured as an average over sixty runs with the same picture on every card and the cursor at its centre, so that only smell differed. A single stop has none of that: the spread between one look and the next is larger than the whole lesson, two untrained coins already differ from each other by more than it, and the eye window is wider than a card, so a stop away from a card's centre reads its neighbours too — between a third and six sevenths of the window was the card being judged over the first paper session. Nothing corrects for any of this, because nothing may stand between the brain and the order. Every stored look now records how much of the window was the card and how much of the brain fired in it.
• Nothing here is advice or a signal. It is a fly brain with a nose, in a room of coins, spending money that does not exist.

The other half WINDOW

The same brain, on Solana. This page only looks: the PumpFun tape the relay already publishes, a Phoenix mark, and Stonkfun’s public listings. There is no key on this origin and nothing here can sign.

• A window, not a control panel. The site never asks for a key and never stores one. There is no prepare, no submit, and no Phoenix order on this origin. FLY_SOL_SECRET does not appear here.
• The tape is the public relay. clawd-ws.fly.dev is the live PumpFun page; this site embeds a native readout of the same /ws frames clawdws.py already parses, plus GET /health for status, clients, totalLaunches and solana.
• Stonkfun is listings only. Newest tokens and launchable pairs come from https://www.stonkfun.xyz/api/public/v1. A launch is two POSTs and a local signature, and that path stays in stonkfun.py, disarmed unless FLY_SOL_LIVE=1.