DIY Fly Brain
(Yes, Really)

Real wiring. Simulated activity.

The MaleCNS project reconstructed the wiring of a male fruit fly’s central nervous system and made the data public. Independent developers have connected a model of that wiring to charts and game frames.

The biological wiring is real data. The dynamics, “eyes,” actions and rewards are engineering choices. This is not a mind upload, and changing synapses is not proof that it learned.

Your first fly: paper-trading Bitcoin

Stonkfly uses real public BTC-USDC prices and a fake $100 balance. No Coinbase key needed.

1. Open a terminal and install

git clone https://github.com/nftechie/stonkfly.git
cd stonkfly
python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[test]'

2. Prepare the brain, then run it

The first command downloads the data and builds the graph. Let it finish before the simulation starts.

python -m stonkfly prepare
python -m stonkfly run

python -m stonkfly status
python -m pytest -q

Setup reference: Stonkfly README. Commands checked against the repo on 14 September 2026; not a claim that this experiment has been run or validated here.

Make it play Doom

Doomfly feeds game frames into the model and maps neural activity to movement, turning and firing. This takes more setup than paper trading.

The authors report that the current v6 candidate failed its visual, conditioning and survival validation gates. Treat it as an experiment, not a trained game-playing brain.

git clone https://github.com/nftechie/doomfly.git
cd doomfly
python3.11 -m venv .venv-neural
source .venv-neural/bin/activate
python -m pip install --upgrade pip
python -m pip install -r requirements-neural.txt -r doom/requirements.txt \
  --build-constraint neural-build-constraints.txt

python - <<'PY'
from pathlib import Path
import hashlib, json, urllib.request

name = 'malecns_v1'
registry = json.loads(Path('doom/datasets.json').read_text())['datasets'][name]
locked = json.loads(Path(f'data-provenance/{name}/source.lock.json').read_text())
root = Path('connectome_data') / name
root.mkdir(parents=True, exist_ok=True)
for filename, url in registry['files'].items():
    target = root / filename
    if not target.exists():
        partial = target.with_suffix('.download')
        urllib.request.urlretrieve(url, partial)
        partial.replace(target)
    with target.open('rb') as stream:
        digest = hashlib.file_digest(stream, 'sha256').hexdigest()
    if digest != locked[filename]['sha256']:
        raise RuntimeError(f'Source checksum mismatch: {filename}')
(root / 'source.lock.json').write_text(json.dumps(locked, indent=2) + '\n')
PY

python -m doom.connectome malecns_v1
python -m doom.prepare
python -m doom.audit_data
python -m doom.build_kernel
python -m doom.server --model experimental-v6 --learning --port 8766 \
  --audit-dir outputs/doom/local-training \
  --checkpoint-dir outputs/doom/local-training/checkpoints \
  --checkpoint-seconds 300 --resume

cd /path/to/doomfly/doom-ui
npm ci
# Create .dev.vars in this folder with the line below:
# DOOM_STREAM_ORIGIN=http://localhost:8766
npm run dev

Same wiring. Your experiment.

Fork Stonkfly or Doomfly and change three adapters. This is a coding project, not a prompt that instantly teaches it a new skill.

Five ideas to borrow

Experiment ideas, not demonstrated capabilities.

• Tweet rankerFeed screenshots. Compare its picks with later engagement.
• Reel topic pickerShow thumbnails and hooks. Test its choices against a simple baseline.
• Paper-market flySwap the chart source. Keep every order simulated.
• Focus experimentUse a local activity summary as input and a focus score as feedback.
• Meme tournamentRun two seeds on the same inputs. Compare their choices.

Start with local logs instead of real-world actions. Compare against random choices and a version without weight updates before calling the result “learning.”

Just want to look inside?

Browse cell types, connectivity and eye maps through the official project. Or query the male-cns:v1.0 dataset in neuPrint.

python -m pip install neuprint-python

from getpass import getpass
from neuprint import Client, fetch_neurons

client = Client(
    "https://neuprint.janelia.org",
    dataset="male-cns:v1.0",
    token=getpass("neuPrint token: "),
)
neurons, syndist = fetch_neurons("DNge104")
print(neurons.head())