Ways to try the fly
Try a fly brain in a computer: online demos and no-install options
You do not need to install anything to see a simulated fruit fly brain at work. We found 41 pages that run in a browser or a free online notebook (one of them is a live stream), and on 3 and 6 Oct 2026 we loaded eleven of the browser demos in a real browser to see whether they actually run. Start with a browser demo, move to a notebook if you want to change something, and download code only when you are ready to run it yourself.
Want to see a demo before you open one? Watch the demos in the Video gallery: 54 videos, each with the project's evidence grade and what the video does not show.
Fly brain simulators you can try online: does it actually run?
A link that answers is not a demo that works. So we open online fly-brain simulators in a real browser, a few at a time, with a procedure written down before the first load: 11 so far, 8 of which ran. None computed its brain on a server.
6 Oct 2026: five more demos, all ran
Vial, Doodle Fly, Infinite Sugar, Fly Dino and Neural Canvas, with the same browser and rules as on 3 Oct. They downloaded between 3.8 and 36 MB in their first minute.
- 5 of 5 ranone load each, no page errors, memory peak 0.74–1.28 GB (our limit: 1.6 GB)
- 3 compute in your browserVial and Doodle Fly in a Web Worker, Infinite Sugar in WebAssembly; for Fly Dino and Neural Canvas our rule could not tell where
- 1 catalogue size not seenNeural Canvas ran, but its ~77 MB brain download was not seen in our check window
Read this before sharing. Checked in headless Chromium 154 on a Linux server without a GPU, 6 Oct 2026. Shows whether the page loads, downloads its brain data and keeps computing; it is not a play test. Safari, Firefox, phones and GPU machines were not tested.
All five ran by our fixed rule, with one load each, no page errors and no server calls; the browser's memory peaked at 0.74–1.28 GB, under our 1.6 GB limit. "Ran" means the page loaded, downloaded data and kept changing; it does not mean we played it. Neural Canvas ran, but its ~77 MB brain download was not seen in our check window, so the catalogue keeps its size figure with that note.
Results of 6 Oct
| Demo and grade | In our browser | Where it computes | Downloaded | Largest file | First movement | Errors | Memory peak | Matches our catalogue? | What we saw |
|---|---|---|---|---|---|---|---|---|---|
| Infinite Sugar B Open the demo |
Ran Our one action: Typed "Fly" into the name field and pressed Enter |
in your browser (WebAssembly) | 36.04 MB | colidx.bin.gz 7.12 MB | 36.8 s | 0 / 0 | 1.28 GB | Yes (36.04 MB vs ~47 MB) | Ran after we typed a name ("Fly") and pressed Enter, as its README says; MuJoCo and the brain run in WebAssembly. 36.04 MB against about 47 MB in our catalogue. |
| Doodle Fly B Open the demo |
Ran Our one action: None documented; it starts by itself |
in your browser (Web Worker) | 34.99 MB | graph.bin.gz 31.48 MB | 12.1 s | 0 / 0 | 0.98 GB | Yes (34.99 MB) | Ran with no action needed; a 31.48 MB brain graph loads into a Web Worker. Our catalogue had no size before this check. |
| Vial B Open the demo |
Ran Our one action: None documented; it starts by itself |
in your browser (Web Worker) | 7.43 MB | conn.bin.gz 6.84 MB | 10.6 s | 0 / 0 | 0.74 GB | Yes (7.43 MB vs ~8 MB) | Ran with no action needed; its 6.84 MB wiring file (conn.bin.gz) loads into a Web Worker. |
| Neural Canvas B Open the demo |
Ran Our one action: Clicked "Download & start" |
unclear | 5.87 MB catalogue: ~77 MB brain download, not seen in our check window |
a Hugging Face CDN file 2.38 MB | 39.7 s | 0 / 0 | 0.84 GB | No: the brain download in our catalogue was not seen | Ran: after we clicked "Download & start" the 3-D view moved and the text changed. Its ~77 MB brain download was not seen in our check window (5.87 MB came down in 60–90 s and no Web Worker started). It may start later than our window, need a GPU path, or the page may have changed; we will re-read the code. |
| Fly Dino C Open the demo |
Ran Our one action: None documented; it starts by itself |
unclear | 3.81 MB | model.bin 1.71 MB | 12.2 s | 0 / 0 | 0.76 GB | Yes (3.81 MB vs ~4 MB) | Ran with no action needed; the game runs on the main page with a 1.71 MB model file and no worker, so by our fixed rule where it computes is "unclear". |
3 Oct 2026: the first six
On 3 Oct 2026 we opened six online fly-brain simulators in a headless Chrome browser (Chromium 154 on a Linux server with 2 CPUs and no GPU), took one action from each README, and recorded what the page downloaded, whether it kept computing and where the brain ran: they downloaded between 2.6 and 34 MB in their first minute. We wrote the procedure down before the first load.
- 3 of 6 ranFly Worker (17.55 MB), FLY67 (33.28 MB) and flybrain.app (12.72 MB)
- 0 on a serverno demo computed its brain on a server; where we could locate it, it ran in a Web Worker in your browser tab
- 2 froze, 1 hit our memory limitmost likely because our machine has no GPU (software WebGL); not tested on GPU machines
Read this before sharing. Checked in headless Chromium 154 on a Linux server without a GPU, 3 Oct 2026. Shows whether the page loads, downloads its brain data and keeps computing; it is not a play test. Safari, Firefox, phones and GPU machines were not tested.
Every 3-D page fell back to software WebGL because our machine has no GPU. The two freezes most likely come from that, not from the demos; on a computer with a graphics card they may run normally. "Stopped at our 1.6 GB memory limit" means our test machine's limit was reached, not that the demo is faulty. A frozen or stopped page may also have downloaded less than it would on a GPU machine.
Results of 3 Oct
| Demo and grade | In our browser | Where the brain ran | Downloaded | Largest file | First movement | Errors | Memory peak | Matches our catalogue? | What we saw |
|---|---|---|---|---|---|---|---|---|---|
| FlyLeno B Open the demo |
Loaded, then froze in our no-GPU browser Our one action: Start without music (the page did not answer) |
in your browser tab (Web Worker) | 33.95 MB brain data 31.36 MB |
connectome.bin.gz 31.36 MB (worker) | 10.4 s | 0 / 0 | 1.04 GB | Unclear: the page did not run to the end (froze) | Its 31.4 MB connectome reached a Web Worker and the counters started; then the 3-D stage stopped answering after about 10 s. |
| fly-brain (Lulzx) A Open the demo |
Stopped at our 1.6 GB memory limit Our one action: Play (clicked) |
unclear | 33.93 MB brain data 14.63 MB |
graph.flyg 14.63 MB (service worker) | not seen in screenshots | 2 / 0 | 1.63 GB | Unclear: the page did not run to the end (memory stop) | Fetched its 14.6 MB MaleCNS graph and WebAssembly and started 4–5 workers, then passed our 1.6 GB browser memory limit on both loads (at 29 s and 44 s). |
| FLY67 B Open the demo |
Ran Our one action: Feed it sugar (worked) |
in your browser tab (Web Worker) | 33.28 MB brain data 31.24 MB |
connectome.bin.gz.part00 15.74 MB (worker) | 11.6 s | 0 / 0 | 1.00 GB | No: our catalogue said about 100 MB (an estimate); corrected to 33.3 MB. The mechanism matches. | Two connectome parts (15.7 + 15.5 MB) load into a Web Worker; the fly moved and the sugar button worked. Our old "~100 MB" estimate was wrong. |
| Fly Worker A Open the demo |
Ran Our one action: Hire a fly (worked) |
in your browser tab (Web Worker) | 17.55 MB brain data 7.55 MB |
carnage3D.data 8.18 MB | not seen in screenshots | 0 / 0 | 1.09 GB | Yes (worker seen; 17.55 MB vs ~13 MB) | Its 7.6 MB brain file runs in a Web Worker and the on-page simulation numbers kept changing; 8.2 MB more is GTA1 game data. |
| FlyBrain (flybrain.app) C reserve, replaced fly-brain (Lulzx) Open the demo |
Ran No documented action |
unclear also tries a chat WebSocket and chat server (both failed here) |
12.72 MB brain data 12.45 MB |
connectome.bin.gz 12.45 MB | 10.9 s | 2 / 0 | 0.87 GB | Yes (worker seen; 12.72 MB vs ~12.4 MB) | Its 12.5 MB connectome loads and a Web Worker runs; the page also tries a chat WebSocket and a chat server (both failed here), so where the brain runs is unclear by our fixed rule. |
| Help the Fly Escape B Open the demo |
Loaded, then froze in our no-GPU browser Our one action: Release the flies (the page did not answer) |
unclear | 2.57 MB its 9.9 MB brain graph never arrived |
index-DzgoIUHH.js 1.56 MB | not seen in screenshots | 0 / 0 | 0.86 GB | Unclear: the page did not run to the end (froze) | Loaded its WebAssembly (1 MB) and a worker, but stopped answering from the first screenshot; its 9.9 MB brain graph never arrived in 84 s. |
| webgpu-fly C extra 20 s record, not the full check |
Did not start: needs WebGPU | not reached | 0.29 MB fonts only |
fonts | none in 20 s | 0 / 0 | 0.72 GB | Yes: the catalogue says it needs a GPU with WebGPU | Chrome logged "No available adapters.", its WebGPU message: our browser has no GPU adapter. Our script's fixed rule called this "loads, no activity". |
- flybrain.app: the page also tries to open a chat WebSocket and a chat server (both failed in our browser), so we cannot say for sure that nothing runs elsewhere; by our fixed rule, where its brain runs is "unclear".
- webgpu-fly needs WebGPU. In a 20 s extra record it loaded 0.29 MB and did not start: Chrome reported no graphics adapter, as expected without a GPU.
- Our correction: our catalogue said FLY67 downloads about 100 MB. It downloaded 33.3 MB, and the entry is fixed.
- Not tested: Safari, Firefox, phones and computers with a GPU. A GPU or a different software-graphics setting would change the test, so it needs its own written plan before we repeat it.
Four levels of effort
- Level 1Open in a browserNothing to install. Watch or play with a demo that runs in the page.Browser demos
- Level 2Run a notebookA free online notebook runs real research code for you. You can change a number and run it again.Notebooks
- Level 3Download and runInstall Python or an app and run a project on your own computer.Download and run
- Level 4Build your ownFollow our guide from an empty folder to your own small project.Build your own
What our check labels mean
- Page respondsThe link answered on the check date (HTTP 200), but we did not load the demo in a browser. It may still need a special browser, a sign-in or a fast computer.
- RanIn our headless browser check (no GPU) the page loaded its brain data and kept computing. Not a play test: we took one action and watched for 60–90 s.
- Loaded, then froze in our no-GPU browserThe page loaded, then stopped answering. Most likely our machine's software graphics (no GPU), not the demo; it may run on your computer.
- Stopped at our 1.6 GB memory limitThe page used more memory than our test machine allows. A computer with more memory may run it.
- DownloadNeeds an install. We give the platform and download size when the authors state them.
A demo that opens is not the same as a demo driven by the fly's brain. Each row also shows the project's evidence grade: A and B mean a real connectome drives the behaviour; C, D and U mean most of what you see comes from other code, or we could not check.
No-install options (levels 1 and 2)
Grouped by what you can do. Inside each group, the best-supported projects come first.
Browser demos where the simulated brain drives what you see (10)
A real connectome runs in the page (grades A and B). People still chose which neurons get the input and which neurons move the fly.
| Name | What it is | Grade | What you need | Our check | Links |
|---|---|---|---|---|---|
| ChessFly Ruben Nugmanov (znatgost; also the author of fly67) Catalogue entry chessfly |
A browser game in which a mushroom-body model learns chess by self-play. The 285 olfactory projection neurons, 5,177 Kenyon cells and their 21,509 connections (at least 5 synapses each) come from FlyWire v783; the board is fed onto the projection neurons through a fixed hand-made mapping, Kenyon cells fire when at least 2 inputs are on, and one output neuron's plastic synapses learn from game results. A small alpha-beta search picks the move. The author trained the real wiring, a shuffled-claws control and the original random design for 10,000 games each. | A | Download: small: brains/*.json about 30 KB each, data/mushroom_body.json (not measured in a browser) | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| Fly Worker hwkim3330 Catalogue entry fly-worker |
A browser tool that plays games with key presses, reads the screen, and records bugs so they can be replayed. One of its steering policies is a spiking model of the whole FlyWire v783 brain in a Web Worker. The author benchmarked it against random-steering baselines, and it does worse. | A | Download: ~13 MB (brain.bin 11.4 MB raw / ~7.45 MB gzip, meta.json 0.48 MB, pos.bin 0.97 MB), plus… | Ran Our browser check, 3 Oct 2026: 17.55 MB downloaded; brain ran in your browser (Web Worker). Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code No licence file README says MIT but no LICENSE file |
| fly-brain Lulzx Catalogue entry fly-brain-lulzx |
Runs the whole MaleCNS v1.0 connectome (165,122 neurons) as a spiking network for each fly in the browser, with a MuJoCo flybody body and a flyvis compound eye. Descending-neuron activity sets walking, turning, grooming and escape commands. Documented offline experiments score the model against literature targets and against shuffled-wiring controls. | A | Faster with a GPU Download: ~23 MB for the arena (README); ~30 MB for the viewer |
Stopped at our 1.6 GB memory limit Our browser check, 3 Oct 2026: 33.93 MB downloaded; brain ran where: unclear. Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code MIT |
| Doodle Fly dtecx Catalogue entry doodle-fly |
The whole FlyWire v783 brain (Shiu et al. LIF) runs in the browser and plays a Doodle-Jump-style game: the target platform's horizontal offset drives the LC10a population of each eye as Poisson input, and the smoothed rates of left and right steering descending neurons (DNa02, DNa01 and others), balanced by a start-up calibration, press the left and right buttons. Two built-in controls (swap eyes, blind) break the play. | B | A recent desktop browser | Ran Our browser check, 6 Oct 2026: 34.99 MB downloaded; brain ran in your browser (Web Worker). Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code MIT |
| FLY67 znatgost Catalogue entry fly67 |
The whole FlyWire v783 brain (138,639 neurons, 15.1 M connections, Shiu et al. LIF parameters) runs in a browser Web Worker and drives an animated 3D fly: sugar under the proboscis drives MN9, bitter suppresses it, an air puff drives antennal grooming via aDN1, and a looming shadow recruits the giant fibre and DNa02. Descending-neuron rates are turned into movement with hand-set gains; a '6-7' meme gesture is a labelled scripted easter egg. | B | Download: 33.3 MB transferred in our browser check (connectome.bin.gz in two parts, 15.7 + 15.5… | Ran Our browser check, 3 Oct 2026: 33.28 MB downloaded; brain ran in your browser (Web Worker). Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code MIT |
| FlyLeno (TUURD Talk) AgitationSkeleton Catalogue entry flyleno |
A browser talk show in which the whole FlyWire v783 brain, run live as the Shiu et al. spiking model in a Web Worker, drives a ragdoll Grey Leno. Show events, music and a thrown sugar cube stimulate chosen sensory neurons; descending-neuron firing rates are mapped by hand to walking, turning, startle, grooming, feeding, voice and body muscles. Gait, balance, food seeking, homing, sleep and the choice of spontaneous actions are engineered layers that the README labels as such. | B | Download: about 32 MB of connectome and position data plus 3-D assets, fetched by the page | Loaded, then froze in our no-GPU browser Our browser check, 3 Oct 2026: 33.95 MB downloaded; brain ran in your browser (Web Worker). Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code No licence file |
| Help the Fly Escape dzhng Catalogue entry help-the-fly-escape |
A 3D browser puzzle game: you place household objects, then a swarm of flies tries to find the exit. Each fly runs its own noisy LIF simulation of a 70,000-neuron MaleCNS subgraph in Rust/WASM. Readouts from descending and motor neurons set each fly's walking thrust, turning and takeoff. | B | Download: ~11 MB brain data (graph.bin 9.9 MB, manifest 0.6 MB, retinal map 0.2 MB) plus WASM and… | Loaded, then froze in our no-GPU browser Our browser check, 3 Oct 2026: 2.57 MB downloaded; brain ran where: unclear. Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code No licence file |
| Infinite Sugar cnqso Catalogue entry infinite-sugar |
A browser artwork: a MuJoCo flybody fly sits in a terrarium while a whole-brain spiking model of FlyWire gets constant sweet-taste input. Activity in proboscis, head and antenna motor neurons moves the matching body joints. | B | Download: 36.04 MB transferred in our browser check (6 Oct 2026; largest file colidx.bin.gz 7.12… | Ran Our browser check, 6 Oct 2026: 36.04 MB downloaded; brain ran in your browser (WebAssembly). Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code No licence file |
| Neural Canvas (Fruit Fly Simulation) Xenova (Hugging Face Space; commit author 'Joshua') Catalogue entry neural-canvas |
You paint neuron positions on a map of the brain, or press walk/turn/fly buttons. The browser then runs a Shiu-style spiking model of the whole MaleCNS v1.0 brain and nerve cord, and descending-neuron firing rates move an animated NeuroMechFly body. | B | Faster with a GPU Download: ~77 MB compressed connectome arrays (sum of manifest parts 76,667,180 bytes) plus body… |
Ran Our browser check, 6 Oct 2026: 5.87 MB downloaded; brain ran where: unclear. Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code MIT app code per LICENSE file; MaleCNS data CC BY 4.0; body and kernels Apache-2.0). Space metadata says apache-2.0 |
| Vial (fly-vial) Raphael Rocha (RaphaelSR) Catalogue entry vial |
A virtual-pet game where you feed, clean and train a 3D fly. A web worker runs a LIF simulation of the whole FlyWire brain, pruned to connections of 5 or more synapses. Firing of chosen descending and motor neurons sets reflex animations, and a small REINFORCE policy learns 'tricks' from descending-neuron activity. | B | Download: 7.43 MB transferred in our browser check (6 Oct 2026; largest file conn.bin.gz 6.84 MB).… | Ran Our browser check, 6 Oct 2026: 7.43 MB downloaded; brain ran in your browser (Web Worker). Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code MIT code); LICENSE notes web/data/ (FlyWire v783-derived) keeps its data licence |
Browser demos with partial, stand-in or scripted brains (10)
Fun to watch, but most of the visible behaviour comes from a small circuit, a synthetic network, scripts or a replay, or we could not check how it works (grades C, D and U).
| Name | What it is | Grade | What you need | Our check | Links |
|---|---|---|---|---|---|
| Fly Dino (flyjump) Mert Cobanov (cobanov) Catalogue entry fly-dino |
Plays the original Chromium Dino game. Eight hand-made game-state numbers feed an 80-neuron subgraph taken from MaleCNS v1.0 and run as a leaky tanh rate model. The activity of 16 descending neurons goes to a trained 243-parameter readout (16-12-3), which picks run, jump or duck. | C | Download: 3.81 MB transferred in our browser check (6 Oct 2026; largest file model.bin 1.71 MB).… | Ran Our browser check, 6 Oct 2026: 3.81 MB downloaded; brain ran where: unclear. Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code Custom licence Cobanov Template Attribution License 1.0 (free use incl. commercial; visible linked credit required; not OSI-approved) |
| FlyBrain snedea Catalogue entry flybrain-app-snedea |
A 2D browser pet fly you can feed, touch, blow air at, or expose to light and temperature. A Web Worker runs spikes over the full FlyWire brain graph, and a panel shows them. What the fly does on screen is chosen by a threshold state machine and hand-coded steering, with the network's group activity as one input. | C | Download: ~12.4 MB (connectome.bin.gz) plus small metadata | Ran Our browser check, 3 Oct 2026: 12.72 MB downloaded; brain ran where: unclear. Headless Chromium 154, Linux, no GPU; not a play test. All browser checks |
Open Code MIT |
| FlyDrones SpikeCalls Catalogue entry flydrones |
A Python package and a browser demo. Camera optic flow drives chosen fly visual neuron types in a LIF network, and six descending-neuron groups are read out as throttle and yaw for a simulated or real drone. The browser demo runs 'MiniFly', an 850-neuron hand-wired stand-in; MaleCNS v1.0 is an optional path in the Python package. | C | Download: ~1.2 GB MaleCNS data (optional; MiniFly needs none) | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| FlyWire Pong zonzujiro Catalogue entry flywire-pong |
A browser page in which a 1,230-neuron network with FlyWire FAFB v783 topology plays Atari Pong (the official ALE emulator in WebAssembly) against the built-in computer player. Selected visual neurons receive a brightness and motion encoding on an 11x11 grid; two descending neurons (DNp15, DNp01) feed a small action decoder. The weights were trained (supervised start, then PPO) in a separate project; the page runs inference only and includes an inspector for every decision. | C | Download: repository about 13.5 MB, including an 11 MB ALE source archive… | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code GPL-2.0 |
| making-fly-play-chess mncrftfrcnm Catalogue entry making-fly-play-chess |
A chess player that turns the board into 782 numbers, pushes them for six steps through a fixed 8,192-neuron patch of the FlyWire v783 wiring (tanh rate units), and scores each legal move with a linear readout trained by self-play on 1,024 randomly chosen cells. A Gradio board lets you play it in Colab, and an optional FlyGym view moves a 3-D fly's joints from readout-weighted activity. A comparison notebook trains the same setup on rewired wiring and on classical features. | C | Faster with a GPU | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code Apache-2.0 |
| Same Smell anzal1 Catalogue entry same-smell |
Shows, side by side, subsets of about 600 neurons of the cVA pheromone pathway from a male (MaleCNS v1.0) and a female (FlyWire v783) brain, spiking in a slowed-down LIF model after a 'puff'. The first descending-neuron spike triggers a puppet animation, but the code fixes the kind of reaction (lunge or pause) by sex. | C | Download: ~11 MB (data/real JSON ~10 MB plus 0.6 MB bundle) | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code No licence file |
| webgpu-fly abgnydn Catalogue entry webgpu-fly |
A WebGPU browser demo: keys stimulate named descending neurons in a whole-brain FlyWire v783 LIF simulation, and the spikes pass by cell-type name into a MANC nerve-cord LIF simulation. The averaged leg motor output scales a hand-written tripod gait on a flybody MuJoCo/WASM body, and by default it also sets the body's velocity directly. | C | Needs a capable GPU (WebGPU) Download: ~315 MB (brain.bin 126 MB, vnc.bin 43 MB, flybody bundle 140 MB, walking policy 5 MB;… |
Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| CyberLarva ChenYvhang Catalogue entry cyber-larva |
A 3D larva crawling in an editable habitat. The online page runs a simple animation: the larva heads for the nearest food with a sine-wave crawl, and no neurons are simulated. A separate local Python/MuJoCo build can load the Winding et al. L1 larval connectome if the user prepares the data. | D | Download: not stated (browser page is small: three.js + ~180 KB GLB); the local build needs the… | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| flybrain.info Lab not stated (flybrain.info, English/Russian site) Catalogue entry flybrain-info-lab |
A section of the flybrain.info site where you pick one of 8 stimuli (sugar, bitter, water, sound, looming, geosmin, light, 'runaway') and watch a 3D cloud of FlyWire neurons flash. The browser plays back spike runs computed offline with the site's own numpy re-implementation of the Shiu et al. model; nothing is simulated live. The same site is also a project directory (competitor listing). | U | Download: not stated (brain.bin plus 8 small replay files) | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open No code repository |
| Fruit fly brain plays Pokémon on a Raspberry Pi 5 (Twitch stream) Leetzerzz (Twitch) Catalogue entry fly-pokemon-pi5-stream |
A live Twitch stream in which a spiking model of the FlyWire FAFB v783 brain is said to play Pokémon (listed under FireRed/LeafGreen) on a Raspberry Pi 5 alongside the mGBA emulator. According to the creator, walls, doors and people are read from game memory and fed as left/right drive to looming-detector populations, and descending-neuron activity is decoded into button presses. Chat can trigger reward or punishment stimulation. No code has been published, so the mechanism cannot be checked. | U | A live Twitch stream, not a demo you control. Not watched by us; no code is published. | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open No code repository |
Run real research code in a free online notebook (4)
Google Colab notebooks run the code on a free cloud computer. You need a Google account. You can change a value and run the cell again.
| Name | What it is | Grade | What you need | Our check | Links |
|---|---|---|---|---|---|
| Drosophila_brain_model (Shiu et al. 2024) Philip K. Shiu et al. (Scott lab, UC Berkeley, with FlyWire) Catalogue entry shiu-drosophila-brain-model |
A leaky integrate-and-fire model of all proofread neurons in the FlyWire adult brain, run in Brian2. You activate or silence chosen neurons by FlyWire ID and read out spike rates of every other neuron; the paper tests these predictions against fly experiments. | A | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| flyvis Turaga Lab (HHMI Janelia) / Lappalainen et al. Catalogue entry flyvis |
PyTorch model of the fly visual system (optic lobe motion pathways). Wiring and synapse signs come from an EM column connectome; unknown neuron and synapse parameters are trained on an optic-flow task, and the model responses are then compared with published recordings. | A | Faster with a GPU | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| connectome_interpreter Yijie Yin et al. Catalogue entry connectome-interpreter |
A Python library (on PyPI) that turns wiring diagrams into testable hypotheses: effective connectivity across several synapses, activation spreading and path finding, sized to run whole-brain (about 140,000 neurons) analyses on a laptop or in Colab. | n/a | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| NAVis navis-org (Philipp Schlegel et al.) Catalogue entry navis |
Python library for loading, analysing, comparing and plotting neuron morphology (skeletons, meshes, dotprops), with NBLAST, transforms between brain templates, and loaders for neuPrint, MICrONS, H01 and others. | n/a | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code GPL-3.0 |
Interactive results and benchmarks (8)
Pages where research projects show their results and control tests. Read these to see how the wiring is tested.
| Name | What it is | Grade | What you need | Our check | Links |
|---|---|---|---|---|---|
| Connectome ping pong (fly tennis) castor639 (Warpfield) Catalogue entry fly-tennis |
Two simulated flies rally a ball on a fly-scaled tennis court. Each runs the MaleCNS v1.0 brain and nerve cord (163,903 neurons, edges with 5 or more synapses) as an untrained flyvis-style rate network. A geometric model of FlyGym's 721-ommatidium eye feeds the medulla columns; a decoder reads the ball's bearing from small-object medulla cells (Mi1, Tm3, T2, T3) and moves the fly along its baseline. Matches are compared with a shuffled-wiring control, a parked fly and a descending-neuron (DNa01/02) decoder. | A | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| Flight-test the fly Mutaqin Aryawijaya (aryawidjaja) Catalogue entry flight-test-the-fly |
A study that runs a 20,556-neuron FlyWire v783 subcircuit (every neuron within 4 hops of the motion-sensing T4/T5 cells to the steering neurons DNa02 and DNg02) with the Shiu et al. 2024 spiking model in Brian2 as the rudder yaw damper of a simulated aircraft (a 2-state Dutch-roll model and the JSBSim Cessna 172). The yaw rate sets T4/T5 Poisson rates; the rudder follows the left-right difference of DNa02 firing. Every result is compared against degree- and sign-preserving shuffles of the wiring, a type-preserving shuffle, the airframe with no controller and a classical yaw damper, with hypotheses fixed in advance. | A | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| Fly Self Driving suanmiao (built with agents in a Kylon workspace) Catalogue entry fly-self-driving |
The traced MaleCNS graph (165,122 neurons, 25.6 M edges) runs as a recurrent rate network. Frozen random maps connect pixels to optic-lobe sensory cells and ventral-cord motor cells to steering. One gain per synapse and one leak per neuron (25.7 M parameters) are trained by imitating a pure-pursuit expert plus DAgger, to drive a ray-cast street with traffic. Code includes a randomly rewired graph control and linear and MLP baselines. | A | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| flybench Brandon Cho Catalogue entry flybench |
A benchmark of 36 cited fly behaviours (reflexes, sparseness, escape, and more) that scores whole-brain spiking simulations of FlyWire v783 and MaleCNS v1.0. Every task can also be run on degree-preserving, random and sign-shuffled wiring, and the score gap is reported as 'specificity'. | A | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
| The Fly's Hash Function realgauravvyas Catalogue entry flys-hash-function |
Tests whether the measured hemibrain projection-neuron to Kenyon-cell wiring works better as a FlyHash similarity-search hash than the random matrix used in the literature, with several matched null matrices. A second study tries a mushroom-body readout on sudoku sub-steps. | A | Faster with a GPU | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code No licence file |
| BioReservoir IG Digital Lab (igdigitallab) Catalogue entry bioreservoir |
Runs the MaleCNS v1.0 connectome as a fixed spiking reservoir (Brian2 LIF, Shiu et al. port). Question text is embedded, projected at random onto sensory neurons, and the left/right bias of descending neurons is read as a yes/no answer. Runs are pre-registered and compared with rewired, random-graph, no-brain and coin controls. | B | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code Apache-2.0 |
| FlyArm yusenthebot Catalogue entry flyarm |
A frozen MaleCNS v1.0 connectome (166,700 neurons, 10.5 million connections, run as a rate network on Apple Silicon with MLX) is the recurrent core of a controller for a simulated Franka arm in MuJoCo. Only an encoder into the 1,846 ascending neurons and a linear decoder from the descending and motor neurons are trained, by imitation and then PPO. It reports multi-step manipulation and FrankaKitchen results, lesions against size-matched random sets, and early degree-preserving shuffle comparisons that the author says do not yet establish a wiring advantage. | B | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code Apache-2.0 |
| Wired Different (ConnectomeLens) Dhruvin Sarkar Catalogue entry wired-different |
A gradient-boosted classifier uses graph and neuropil features of the MaleCNS cell-type graph to rank which cell types are annotated as sex-related. It is tested against 500 degree-preserving randomised wirings. | n/a | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code MIT |
Explore the real connectome data (8)
Official data browsers. You can look up single neurons and their connections. Some need a free sign-in to download.
| Name | What it is | Grade | What you need | Our check | Links |
|---|---|---|---|---|---|
| BANC brain-and-nerve-cord connectome Lee lab (Harvard) and the BANC community; Bates, Phelps, Kim, Yang et al. Catalogue entry banc |
Connectome of one adult female fly's brain and ventral nerve cord in a single animal, so signals can be traced from sensors through the neck to motor circuits. The linked repository holds the paper's analysis code and a metadata snapshot. | n/a | Codex (Google sign-in) | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code No licence file |
| Codex (FlyWire Connectome Data Explorer) Murthy Lab, Princeton Catalogue entry flywire-codex |
Flask web app behind codex.flywire.ai for searching neurons, cell types, connectivity, pathways and statistics in the FlyWire whole-brain connectome. | n/a | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code Apache-2.0 |
| FlyWire FAFB whole-brain connectome FlyWire Consortium (Princeton: Murthy and Seung labs; Janelia; Cambridge and others) Catalogue entry flywire-fafb |
Proofread wiring diagram of one adult female fruit fly brain, including both optic lobes but not the nerve cord. It is the data behind the Shiu whole-brain model and most FlyWire-based demos. | n/a | Codex explorer; Google sign-in required for apps and downloads | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open No code repository |
| Hemibrain connectome Janelia FlyEM and Google Connectomics Catalogue entry hemibrain |
Earlier dense connectome of a large part of one female fly's central brain (no optic lobes, no nerve cord), including the mushroom body and central complex. Still widely used for learning and navigation circuits. | n/a | neuPrint explorer | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open No code repository |
| Larval L1 brain connectome (Winding et al. 2023) Winding, Pedigo, Barnes et al. (Cambridge, MRC LMB, Janelia, Johns Hopkins) Catalogue entry larva-l1 |
Complete synapse-level wiring of the brain of a first-instar fruit fly larva: about 3,000 neurons, small enough to run whole on any laptop. It is the usual choice for reservoir-computing experiments. | n/a | CATMAID (L1 Larval CNS) per the paper's data statement | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open No code repository |
| MaleCNS connectome (male central nervous system) Janelia FlyEM with Cambridge (Zoology), MRC LMB and Google Research Catalogue entry malecns |
Proofread connectome of one adult male fly's whole central nervous system: brain, optic lobes and ventral nerve cord joined through the neck. Its September 2026 paper set off the wave of game and meme projects. | n/a | neuPrint explorer (account needed for API token); also Codex MCNS v1.0 | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open No code repository |
| MANC male adult nerve cord connectome Janelia FlyEM, Cambridge Connectomics Group and Google Research Catalogue entry manc |
Dense connectome of the ventral nerve cord of one adult male fly, the part that runs legs, wings and most motor output. Used for walking and descending-neuron studies. | n/a | neuPrint explorer (account needed for API token); also Codex MANC v1.2.1 | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open No code repository |
| Virtual Fly Brain (VFB) Virtual Fly Brain consortium (University of Edinburgh, University of Cambridge and partners) Catalogue entry virtual-fly-brain |
An online atlas and query hub for Drosophila neuroanatomy that brings light-microscopy images, neuron types and connectome data from several datasets into one browsable, searchable site. It offers a 3D browser, Python and R clients, and a Model Context Protocol (MCP) server so that LLM tools can query it. | n/a | A recent desktop browser | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code several: VFB3-MCP MIT VFBquery Apache-2.0; database and ontology repositories CC-BY-4.0 (per GitHub licence detection |
Look at a simulated fly body (1)
A body model without a brain: the controllers are written by the user.
| Name | What it is | Grade | What you need | Our check | Links |
|---|---|---|---|---|---|
| FlyGym (NeuroMechFly v2) Neuroengineering Laboratory (Ramdya lab), EPFL Catalogue entry flygym |
Python library for NeuroMechFly, a MuJoCo body model of the adult fly with vision, smell, leg adhesion and joint sensing. Users compose flies and scenes and plug in their own controllers; version 2 also offers GPU batch simulation through MuJoCo Warp. | n/a | Faster with a GPU | Page responds HTTP 200, 6 Oct 2026. Not played by us. |
Open Code Apache-2.0 |
If a demo stays blank: the demos we checked downloaded 3 to 34 MB in their first minute, and some need WebGPU or a graphics card. Try a recent desktop browser, wait for the download to finish, and read the demo's own notes. Phones and older laptops may not run the larger models. The Codex and neuPrint data browsers ask you to sign in for some features.
Download and run (level 3)
These games, desktop apps, brain engines and bodies run on your own computer. Many need gigabytes of connectome data, and some need an NVIDIA GPU or a Mac. We have not run them in this release; the facts come from their code and documentation. Research code and data tools are in the Projects catalogue.
| Project | Kind | Grade | Runs on | Download | Code and licence | Last upstream activity |
|---|---|---|---|---|---|---|
| Open Fly Entry open-fly |
Browser demo | B | browser page with a WebAssembly game; Colab notebook for the Brian2 path GPU: none |
Shiu v783 files (about 100 MB) fetched by the workflow or notebook | Code Apache-2.0 |
|
| Fly x Jev Entry fly-x-jev |
Browser demo | C | browser front end plus a local Python 3.9+ server that proxies the TypeSafe API GPU: none |
repository with prebuilt arena/data/ (size: see gh_meta) | Code MIT |
|
| Brain Runners Entry brain-runners |
Game | A | Python 3.12 (uv) for the fly; browser front end for viewing GPU: none |
about 105 MB Shiu et al. data plus Python dependencies | Code MIT |
|
| Digital Fly in Terraria (MaleCNS fly with a body) Entry fly-terraria |
Game | B | Terraria with tModLoader (Windows .cmd launchers); native C++ runtime, optional CUDA GPU: optional |
repository plus a 131 MB Git LFS synaptic checkpoint; MaleCNS feather files downloaded separately | Code MIT |
|
| DOOMFLY Entry doomfly |
Game | B | Linux, macOS GPU: none |
MaleCNS v1.0 tables (several GB RAM; download size not stated) | Code MIT |
|
| Fly Brain Minecraft Entry fly-brain-minecraft |
Game | B | Windows, macOS, Linux GPU: none |
~23 MB connectome inside the mod jar (README); build from source with JDK 21 (no release jar) | Code MIT code); LICENSE appends a data note: male-cns:v1.0 data CC BY 4.0 |
|
| Fly-NAF Entry fly-naf |
Game | B | Windows, or Linux on X11 with the game under Wine; needs Five Nights at Freddy's 1 GPU: CUDA optional; about 3.6 frames per second on CPU (README) |
FlyWire v783 files fetched separately; needs the game | Code GPL-3.0 |
|
| Fly64 Entry fly64 |
Game | B | macOS GPU: none |
~1.1 GB brain data, plus a user-supplied SM64 ROM and sm64ex build | Code No licence file |
|
| FlyArena Entry flyarena-banc |
Game | B | Windows GPU: required |
not stated (BANC 888 files downloaded at setup) | Code Custom licence FASL-1.1 (view and run for non-commercial use; modified redistribution or commercial use need the author's written permission) |
|
| FlyBrain · Flappy (escape reflex) Entry flybrain-flappy |
Game | B | Python 3.10 or newer with numpy, pandas and PyTorch (CPU is enough); launchers are Windows .bat files; the server itself is standard-library HTTP GPU: optional |
repository about 53 MB at HEAD, mostly the 50.5 MB whole-brain asset data/spiking_full.npz | Code No licence file |
not known |
| Flyhard (The Driving Fly) Entry flyhard |
Game | B | Linux GPU: required |
~1.1 GB MaleCNS weights table plus CARLA 0.9.16 runtime | Code MIT |
|
| FlyKart Entry flykart |
Game | B | local Python server opened in a browser; Windows via WSL2; macOS runs on CPU in slow motion GPU: NVIDIA GPU for real time; since 2026-09-29 an accelerated CPU path (docs/cpu-performance.md) runs without one, more slowly |
about 1.2 GB connectome download; about 8 GB disk including PyTorch/CUDA | Code MIT |
|
| Stonkfly Entry stonkfly |
Game | B | macOS, Linux GPU: none |
MaleCNS v1.0 tables, 'several GB' (README); 16 GB RAM recommended | Code MIT |
|
| fly-plays-games (Pokémon Red chapter; formerly fly-plays-pokemon) Entry fly-plays-games |
Game | C | Python with PyBoy on a laptop CPU; setup uses Unix shell commands; Windows not stated GPU: FlyBrain accepts --device cpu, cuda or auto; README says it runs on a laptop CPU |
brain.npz and weights.npz about 260 MB on first use, plus the fly.ai repository; Pokémon Red ROM must be supplied by the user | Code MIT |
|
| flybrain (Game Boy fly) Entry acamilo-flybrain |
Game | C | Node.js library (tests and a noise-frame example run in plain Node) plus a Rust service and Linux stream infrastructure; no Raspberry Pi build found GPU: CPU kernel by default; optional CUDA LIF kernel in the Rust service (flybrain-core) |
connectome artifacts in repo about 11 MB (data/fafb-v783); ROM not included | Code Apache-2.0 |
|
| FlyBridge Entry flybridge |
Game | C | Windows GPU: required (NVIDIA CUDA for training) |
repo about 70 MB with checkpoints and demonstrations; Minecraft Java 1.21.4 must be owned separately | Code MIT |
|
| FlyCNS Tic-Tac-Toe Entry flycns-tictactoe |
Game | C | Python command line GPU: none |
small; needs a neuPrint token to fetch the sub-circuit | Code MIT |
|
| FlyCraft Entry flycraft |
Game | C | Windows GPU: required |
~600 MB weight cache built on first run (README) plus CUDA PyTorch | Code GPL-2.0 |
|
| flypoker Entry flypoker |
Game | C | local server: Docker, or Python 3.10+ with pip install -e . GPU: none |
repository about 3 MB of data (circuit.npz 1.6 MB uncompressed arrays, brain.stl mesh) | Code No licence file |
|
| FlyWireGBA Entry flywire-gba |
Game | C | Game Boy Advance ROM: mGBA emulator (any desktop OS) or a GBA flash cartridge GPU: none |
ROM 16 MiB (16,777,216 bytes, docs/validation.md) | Code MIT |
|
| Fruit fly vs Jev: Roblox maze race Entry fruitfly-roblox |
Game | C | Python server plus Roblox Studio; the Studio place file is not included, only the scripts GPU: the brain runs on CPU (device='cpu') |
about 260 MB of connectome data downloaded by the flybrain package on first launch; LLM opponent needs a paid OpenRouter key | Code MIT |
|
| Kick the Fly Entry kick-the-fly |
Game | C | Windows, Linux GPU: none (NumPy; optional Numba/torch) |
about 90 MB (Windows exe) / 110 MB (Linux AppImage), plus connectome tables on first build | Code MIT |
|
| FlyBrain-HalfLife Entry flybrain-halflife |
Game | D | Windows, Linux GPU: optional (PyTorch CUDA; CPU lite mode) |
repository about 15 MB; no connectome download needed | Code MIT |
|
| NeuroCraft Fly Entry neurocraft-fly |
Game | U | not stated GPU: not stated |
not stated (only ~32 MB of demo media available) | Code MIT |
|
| DesktopFly Entry desktop-fly |
Desktop app | C | macOS, Windows GPU: none |
~1.2 MB repo tree incl. data (clone metadata tree_bytes) | Code MIT code); LICENSE adds that data/ is CC BY-NC 4.0 (FlyWire-derived) and data/DATA_LICENSE.md puts the MaleCNS extract under CC BY 4.0 |
|
| pianist-fly Entry pianist-fly |
Desktop app | C | Linux GPU: none |
~1.1 GB MaleCNS v1.0 data (not stated exactly) | Code MIT |
|
| FlyBrain Robot Bridge Entry flybrain-robot-bridge |
Desktop app | D | Linux, macOS, Windows GPU: none |
~3 MB repository (mostly images); no connectome download | Code MIT |
|
| Fly With Me (Fly Brain DJ) Entry fly-with-me |
Art | B | local Python server opened in a browser GPU: none |
about 400 MB of connectome data on first run (README) | Code Apache-2.0 |
|
| Bad Apple Fly Entry bad-apple-fly |
Art | C | macOS, Linux GPU: none |
~1 GB connectome tables (README) | Code No licence file |
|
| neurafly Entry neurafly |
Art | C | Rust (stable), Linux desktop with PulseAudio or PipeWire, Unicode braille terminal GPU: none |
18 MB connectome file (data/flywire_net.bin) plus the binary | Code MIT |
|
| FLYBRAIN Bad Apple x DOOM Entry flybrain-bad-apple-doom |
Art | D | local Python server and WebGL browser viewer GPU: WebGL browser |
about 800 MB FlyWire download and 460 MB of processed indices; about 3 GB free disk (README) | Code No licence file |
|
| drosophila-brain-mlx Entry drosophila-brain-mlx |
Brain model or engine | A | macOS GPU: required |
~90 MB upstream FlyWire v630 files + ~114 MB compiled pack; ~1.1 GB for MaleCNS | Code MIT |
|
| dotFly Entry dotfly |
Brain model or engine | C | Windows, Linux GPU: none |
~300 MB FlyWire v630 files, or ~3 GB MaleCNS v1.0 plus checkpoints | Code GPL-3.0 |
|
| fasterfly Entry fasterfly |
Brain model or engine | n/a | Linux GPU: required |
not stated (needs MaleCNS v1.0 feather tables, traced-only weights) | Code No licence file |
|
| fly-brain Entry eon-fly-brain |
Brain model or engine | n/a | Linux GPU: required |
~100 MB connectome data in repo; ~580 MB weight caches generated; spike bundle (600 parquet files) on Google Drive | Code GPL-2.0 |
|
| flybody Entry flybody |
Fly body | n/a | Linux GPU: optional |
not stated (optional figshare data) | Code Apache-2.0 |
For the full list with filters, open the Projects catalogue and tick "Try in browser, no install" or choose a platform.