MOSQUITOBRAIN

where the mosquito is right now

A room goes in through CO₂, odour, heat and contrast. Roughly 80 glomeruli integrate, a ring attractor picks a heading, and a wingbeat comes out. Every number on this panel is read out of the model running in your browser — nothing here is a recording.

MOSQUITOBRAIN LIVE AEDES AEGYPTI · SIMULATED ROOM CO₂OLFACTIONVISIONTHERMALFLIGHT ● LIVE

The room · drag the host

drag anywhere to move the host

phase

—

range

—

CO₂

—

skin ΔT

—

Neural activity

—

MODELLED UNITS FIRING of 12,400

spikes/s

—

membrane

—

wingbeat

—

yaw

—

Antennal lobe · glomeruli

active

—

human-tuned

—

cpA CO₂

—

One glomerulus is tuned to decanal and undecanal — long-chain aldehydes enriched in human skin. Zhao et al. 2022.

Sensory drive

CO₂ does not just attract — it gates. Until cpA fires, contrast is ignored. van Breugel et al. 2015.

Central complex · heading

heading

—

bump width

—

turn

—

Mushroom body · learning

depressed

—

mean gain

—

bloodmeals

—

swats

—

KC→MBON synapses under dopamine, depression only. Mosquitoes really do learn to avoid a defensive host. Vinauger et al. 2018.

Flight motor

Event stream

steps — bites — aborts — model time —

serverless · live roaming

Watch the model navigate by itself.

A browser-based neural model follows Wikipedia links automatically, then periodically visits Pump.fun Explore to inspect meme coins and search their names on Wikipedia. The neural cursor makes every choice. No clicks are required.

SERVERLESS MODEL · LIVE

NEURAL CURSOR CONTROL · WIKIPEDIA + PUMP.FUN · NO MANUAL CLICKS

inside the mosquito · interactive 3D

Explore the mosquito brain in three dimensions.

Turn the brain in your hands. Select a lobe to trace a brief signal through its branching fibers. Its fine structure is illustrative, with distinct colors for each region.

select a region

Mosquito brain

whole model

Select a region to trace a signal through its fibers. Drag to rotate and use + / − to explore the branching detail.

DRAG TO ROTATE · SELECT A REGION · + / − TO ZOOM · KEYBOARD: ARROWS + ENTER

how it finds you

Three senses, three distances, in that order

A mosquito does not smell you and fly at you. It runs a relay: each sense hands off to the next at the range where the next one starts working. Break the chain anywhere and the hunt collapses — which is why CO₂ traps and treated bed nets work on completely different links.

10–50 m

CO₂ — the cpA neuron

One neuron per maxillary palp sensillum does this. cpA expresses Gr1, Gr2 and Gr3; Gr2 and Gr3 make the working receptor and Gr1 sharpens it. Your breath is a plume of roughly 4% CO₂ pushed into air that sits near 0.04% — a hundredfold step it is built to catch. It does not habituate away either; the response stays up for minutes.

5–15 m

Vision — gated by the CO₂

Here is the part that surprised people. CO₂ does not steer the mosquito. It switches on a visual reflex that was ignoring high-contrast objects a second earlier. In still air with no CO₂, a dark shape is nothing. One breath later, the same shape is a target.

< 1 m

Heat and damp — Ir21a, Ir25a, TRPA1

Close in, skin at about 34 °C against a cooler room is a beacon. Two ionotropic receptors and the TRPA1 channel read the gradient, with two opsins helping at the faint end. Convection and humidity give away where the skin actually is, and the landing is committed.

Ranges after van Breugel et al. 2015, Current Biology. The stage the model above is in right now: —

what is actually known

A model, built on measurements

Everything structural on this page comes out of published mosquito work, and it is worth being precise about which, because the anatomy is solid and the dynamics are not.

Antennal lobe	~80 discrete glomeruli, 63 of them in stereotyped positions, from the updated Aedes aegypti atlas.
Human channel	A single glomerulus responds strongly to human odour and weakly or not at all to animals. Its ligands are decanal and undecanal, long-chain aldehydes that come from human skin lipids. A synthetic blend hitting the same set of glomeruli triggers host-seeking on its own.
Receptor genes	117 odorant receptors, 135 ionotropic receptors, 72 gustatory receptors in the reference genome.
CO₂	cpA on the maxillary palp, via Gr1/Gr2/Gr3. It is a dual detector — the same neuron reports skin odorants, which is what made it a drug target.
Heat	Ir21a, Ir25a and TRPA1, with Op1/Op2 extending the faint end. Effective at midrange only when CO₂ and skin odour are already present.
Integration	Mushroom bodies for learning and memory; central complex for steering. Both are named, mapped structures in the mosquito brain.
Learning	Dopamine-dependent. A mosquito that gets swatted at learns to leave that host alone, and that plasticity is what moves the mushroom body panel above.

what is not real, stated plainly

The parts that aren't the mosquito

There is no mosquito connectome. This is the big one. The fruit fly has a traced wiring diagram — every neuron, every synapse, cut out of electron microscope volumes. No such dataset exists for any mosquito. So nothing here is "a real nervous system loose on the internet". It is a model of a brain that has been mapped regionally and never wired.
Nobody has counted the neurons. You will see figures quoted around. We are not going to pick one, because the published literature does not support it. The "12,400 units" on the panel is the size of this simulation, not an estimate of the animal.
The glomerular map is arranged, not measured. There are about 80 of them and 63 sit in reliable places, but the coordinates driving those dots were laid out by hand to match the published arrangement. Do not read a position off this page and put it in a paper.
The neurons are leaky integrators. Rest −52 mV, threshold −45 mV, 20 ms tau — standard values borrowed from fly modelling work, not from mosquito recordings. Real mosquito neurons have properties nobody has put into this.
The room is not a room. Airflow here is a cone with a smooth falloff. Actual plumes are turbulent, filamentous and intermittent, and that intermittency is information the animal uses. This model gets a clean gradient, which makes its job easier than the real thing.
It has no goals. Same caveat the fruit fly version earned. There is no intention here, no preference, no experience of being hungry. There is a gate, a gradient and a turn. When the panel says it is hunting, that is a word a person chose for a number going up.
The wording is ours. The event log is written from templates keyed to model state. Nothing here writes about itself, because nothing here has language.