SPIDERBRAIN

live

What the spider is doing right now

The spider is loose on pons. On this site's server a rig holds a real browser open on the launchpad while the brain runs next to it: whatever the browser shows falls on the principal-eye retina, the model integrates it, and whatever the leg pools send out moves a real mouse. It stalks across cards, pounces on tokens, scrolls the list. It lives in thirty-four-second lives, each picking up where the last one stopped, and only while someone is watching. Every figure on this panel came out of the brain that produced the frame in front of you.

SPIDERBRAIN LIVE JUMPING SPIDER · LOOSE ON PONS VISION MOTOR SIMULATION INTERNET WALLET CAPTURE 6,487 NEURONS 40 NEUROPILS ● LIVE

what the spider is looking at on pons, frame by frame

——

Brain activity

—

CELLS FIRING NOW of 6,487

spikes/sec

—

membrane

—

visual

—

motor

—

Cells firing · where they sit in the brain

Leg pools · the motor side

turn L0

turn R0

stalk L0

stalk R0

freeze0

jump0

Principal eyes · what the retina sees

cells on

—

cells off

—

columns

—

Arcuate body · what it has learned

depressed

—

mean gain

—

rewards

—

punishments

—

Mushroom-body → arcuate-body synapses, weakened by an error signal after a jump that caught nothing. They only ever get weaker — a rule borrowed from a cerebellum, which a spider does not have.

Legs

turn L0

turn R0

stalk0

freeze0

jump0

Log

pages—

jumps—

scrolls—

brain steps—

vetoed—

No wallet, no keyboard, nothing downloaded. The pointer is real and so are the clicks: a jump onto a token card opens it, onto a tab switches it, and past the bottom of the screen scrolls the page. Anything that would commit — Connect, buy, sell, launch — gets logged as what it would have done and vetoed before it lands. Flip the rig to light mode and watch how little it manages to find: same eyes, same wiring, blind. If the panel says the spider is between lives, the server is booting the next browser; if it stays quiet, the rig has spent its monthly budget and the spider is back in its silk retreat.

what it is

A nervous system, not a metaphor for one

This is a spiking model of a jumping spider's brain built on the plan of Su 2026 — An embodied spiking model of visually guided orienting in a jumping spider, with a provenance-tracked synthetic connectome. 6,487 leaky integrate-and-fire neurons in 40 named neuropils, two visual pathways, six motor pools, and the wiring between them laid out the way the salticid synganglion is laid out. It is not a neural network "inspired by" a brain; it is a wiring diagram integrating.

A jumping spider is the animal that looks back at you. A brain of roughly a millimetre, fused into one synganglion, carries eight camera eyes: a pair of forward-facing principal eyes with telephoto optics and an acuity approaching a primate's, whose retinas the spider sweeps across the scene with six tiny muscles, and three pairs of wide-field secondary eyes that act as motion detectors. Land showed fifty years ago that the orienting turn is open-loop: prey moving in a secondary eye's field sets the amplitude of a saccade that swings the principal eyes onto it, then the spider stalks, then it jumps. The plan here is the paper's, after Steinhoff et al.: the principal eyes feed their own first- and second-order neuropils and the arcuate body, the unpaired midline centre that steers the legs; the secondary eyes feed the mushroom bodies; both reach the descending neurons. Each cell runs as a leaky integrate-and-fire unit: rest −55 mV, threshold −45 mV, 20 ms time constant. Excitatory cells excite, inhibitory cells inhibit. The paper's model turns the right way in 99% of trials across fourteen stimulus positions, with turn amplitude proportional to where the stimulus was — Land's measurement, reproduced. That is the behaviour this page inherits.

how it reaches the world

A spider's eyes in, a spider's legs out

The screen falls on a 20×15 field of photoreceptors. The central columns are the principal eyes: a sustained channel that answers contrast against its surround, Weber-fashion, relative to the light it has adapted to. The periphery is the secondary eyes: ON and OFF transients that answer change and fall silent for a scene that holds still. Motion out at the edge drives a turn toward it, scaled by how far out it was; a target held on the principal eyes drives the stalk, and a target centred and close drives the jump. A hunger variable scales the approach. The wiring is never touched. Only synaptic gains change — you do not get to invent a connection the spider does not have.

turn L / R

A pair of descending pools that inhibit each other, so only one speaks. Fed by the secondary eyes' motion channels, weighted by eccentricity: the further out the thing moved, the harder the turn. A jumper swings its whole body to bring it onto the principal eyes.

stalk

The slow creep a jumper does toward prey it has fixed on. Driven by the principal-eye neuropils once a target is on them; the forward pools recruit their own stop.

freeze

Jumpers move in short runs with a still moment between, when the retina re-scans. When the freeze cells fire, the spider holds.

jump

The pounce: dragline fixed, pressure up in the prosoma, legs extend. A jumper commits when this fires, which is why it became the click.

the link to pons

How a spider reaches a launchpad

A spider cannot open a browser, so a rig handles the optics and the spider handles everything after the retina. Four parts, and the log says which of them did what.

the rig · eye and hand

A function on this site's server opens ponsfamily.com/launchpad in a headless browser with the brain running alongside. Several times a second it hands the retina a frame, moves the pointer to wherever the leg pools put the spider, and scrolls to keep up when the spider walks past the edge. It makes no decisions: optics and a pointer, nothing more.

the spider · brain

The frame passes through the principal-eye retina into its neuropils and the arcuate body; the leg pools turn it into stalks and turns, and the jump into clicks. A jump onto a card opens that token. A jump onto a box focuses it — the description and the ticker will be the spider's to fill on launch day. Whatever it catches weakens the synapses that led there, and those gains travel into its next life.

the fence

Connect, buy, sell, launch, approve: any control whose label would commit to something is vetoed the instant the spider lands on it. A token's own website or X link it may follow — it looks for a few seconds, then it is walked back to pons on its dragline. Nothing is ever downloaded and the browser holds no wallet, so nothing could go through even if a veto failed. The name, the pair and the tax are the rig's to type on launch day, and the log will say so.

the click

$SPIDERBRAIN went live on pons in September 2026 — launched by hand, while the spider was busy hunting the launchpad. The address in the bar above is read back from Robinhood Chain, not typed in. The spider's own jump at a launch button, from a wallet of its own, is the thing it gets next: one signed transaction, or none.

the plan

How it got this far

Not a roadmap of promises. A sequence, and everything above the amber line is already built and running on this page.

Connectome loaded6,487 neurons in 40 neuropils on the plan of Su 2026, signed by cell type, with the principal-eye and secondary-eye pathways kept separate

done

Eyes wiredPrincipal eyes as a sustained contrast channel, secondary eyes as ON/OFF motion transients, fed by a real browser on pons through the rig

done

Legs wiredThe turn pair, the stalk pools, the freeze cells and the jump driving a real pointer on pons: hover, scroll, click

done

Tuned without touching the wiringOnly synaptic gains, by cell type, set in a headless harness: a bright patch on the left turns it left, one in the centre draws it on and makes it jump, a dark field leaves it still. No connection was added or removed

done

Reward loop closedMushroom-body → arcuate-body synapses weakened under an error signal, the gains carried from each life into the next

done

Loose on ponsA browser, no schedule, no instructions. A life boots whenever someone is watching and the next begins where the last one ended. Nobody chooses where it goes

running

Token launched on pons$SPIDERBRAIN, September 2026, Robinhood Chain, paired with GOOGL — launched by a person, not by the spider. Contract, name, symbol and supply are read back from the chain below

done

Its own jumpThe veto lifts for a single pounce at a launch button, from the spider's own wallet: the spider types the description and the ticker, the rig the name, the pair and the tax

The other six eyesThe lateral eyes see motion, not detail. Price movement fed in as if something had moved at the edge of vision — a second opinion that is not the principal eyes'

Contract	—
Paired	—
Creator tax	—
Supply	—
Chain	Robinhood Chain · id 4663

what is not real, said plainly

The parts that are not the spider

It is a synthetic connectome. The fly page runs a single male's connectome, traced synapse by synapse. Nobody has done that for a jumping spider, and the paper this follows says so in its own words: the connectome is generated, not measured; no coordinate, count or weight in it is a biological quantity. What is real is the plan — the neuropils, the two eye pathways, the signs of the cells — and the behaviour it reproduces. The paper tags every parameter as measured, literature-derived, estimated or synthetic; this page does not carry those tags, and you should assume "synthetic" for anything it does not name a source for.
6,487 is the paper's number, not a count of a spider. It is the size of the synthetic network. A real salticid brain has not been counted.
It cannot read. Its eyes are a 20×15 patch of cells sensing contrast and change. Text is texture to it: it jumps at a bright heading as readily as a link, and on a form it goes for whichever box it can see, not the right one.
Light mode blinds it. It was tuned against dark interfaces. Switch the capture above to light mode and count what it finds — the fly's developer measured 0 of 3 fields in light and 2 of 3 in dark, and these eyes fail in the same way.
It has a cerebellum's learning rule and no cerebellum. Spiders do not have one. The weaken-only rule is borrowed and applied to the synapses into the arcuate body because that is the region that steers the legs. The direction of the rule is real; its address is a choice.
It lives in thirty-four-second lives. This site's server can only keep a browser open for so long, so the spider lives, dies, and the next life boots where the last one stopped, carrying its counters and its learned gains. Lives run only while someone is watching, and the free tier has a monthly budget; once it is spent, the panel falls back to the spider in its retreat until the month turns.
The rig is a pointer, not a person. It moves the mouse and clicks where the spider tells it to. It never types; on launch day it will fill in the name, the pair and the tax, and the log will mark that as the rig's work rather than the spider's.
It did not launch this token. A person did, on pons, while the spider was roaming the launchpad. The spider's own jump at a launch button is on the list above, not behind it — claiming otherwise would be the one lie on this page.
It has no goals. A spider's brain has no language and no plan. It cannot read a page, hold an opinion about a coin or write anything down. Letting it loose means nobody is steering, not that it decided anything.
The reward signal is made up. What rewards a jumper is a fly, not a web page — a catch stands in for it here, and that is a choice a person made.
It roams inside a fence. It lives on pons and may peek at a token's own website for a few seconds before it is walked back; downloads are denied and there is no wallet. That is a decision, not a limit of the spider: one environment variable takes the fence down, and an unfenced random clicker on a public stream is a bad idea for reasons that have nothing to do with the brain.