Live on Solana and Robinhood Chain

A Fruit Fly Is Learning To Trade. 139,255 Neurons. Real Synapses. Real Money. Solana. Robinhood Chain. Every Minute. CUDA. ROCm. Metal. Any CPU. Open Source. Open Brain. No Stopping.

The complete adult Drosophila connectome, trading graduated memecoins on Solana and Robinhood Chain with real money, and learning from every trade it makes.

01 · About

One brain, two chains, one minute at a time

fly-trader does not use a neural network shaped like a brain. It uses the brain: the synapse-level wiring diagram of an adult fruit fly, run on its real, signed connections, with a mushroom body that keeps learning from what every trade earns.

The brain

The wiring comes from FlyWire FAFB v783, the first complete synaptic connectome of an adult animal brain. Every neuron is present. Every connection between them is a real, electron-microscopy-reconstructed synapse.

Neurons
139,255

Whole brain, both hemispheres, optic lobes included

Connections
2.70 M

≈33.9 M individual synapses

Kenyon cells
5,177

The mushroom body, the fly's learning centre

Output neurons (MBONs)
96

Approach and avoid readouts of the mushroom body

Dopaminergic neurons
323

Reward and punishment signals

Sensory input neurons
5,460

Where the market enters

Descending neurons
1,303

The fly's motor output

Source
FAFB v783

FlyWire consortium, Nature 2024

How it works

Sensing. Once a minute, every graduated memecoin on Solana and Robinhood Chain becomes one row of 68 inputs: returns, order flow, volume, liquidity, age, the token's launch, its creator's record and what winning wallets are doing in it. The row enters the fly through its sensory neurons, the ones that in a living fly carry smell, taste, touch and temperature.

The teacher. A gradient-boosted selector predicts each row's return after every fee and the price impact. It is trained on both chains' history and goes to work only after beating random picks on weeks it never saw.

The fly. Activity flows through the connectome only where the wiring allows. The fly starts as its teacher and buys what the teacher buys; its mushroom body then learns, trade by trade, where the teacher is wrong.

Learning. When a trade closes, dopamine-gated plasticity at the Kenyon cell → MBON synapses, the site of associative learning in a real fly, learns what it made.

Acting. Each buy is sized by how certain the model is and sold when its hold ends.

The trading brain · central brain
Neurons
41,756
Signed connections
1,039,659
Market inputs
68 per row
Plastic synapses
21,438 KC→MBON
Scoring
Every minute
Compute
CUDA · ROCm · Metal · CPU

Training data

The fly learns from the life of memecoins after they graduate: every pump.fun trade since April 2026, and every Pons launch and swap on Robinhood Chain.

pump.fun trades
3.56 B

Since April 2026

Graduated pump.fun tokens
127,215

With their launch and creator history

Pons launches
674,505

Robinhood Chain, since August 2026

Robinhood Chain swaps
41 M

Indexed by fly-trader

What we taught the fly

UNIVERSE

Graduated memecoins, two chains

  • Solana: pump.fun graduates on PumpSwap
  • Robinhood Chain: Pons graduates on Uniswap v4
  • Only tokens with real liquidity and recent volume
STRATEGIES

Fitted, not hand-picked

  • Each strategy has its own model, buy line and hold
  • A win filter and a dump veto screen every pick
THE GATE

Nothing trades unproven

  • Settings are chosen on one stretch of history and judged on a later one
  • A model must make money after costs and beat random picks
  • The fly must pass a replay on each chain before it trades
LEARNING

What the fly may change

  • Only its Kenyon cell → MBON synapses, by dopamine-gated plasticity
  • Its learning can veto a buy, never add one
SIZING

Bet what the record supports

  • A quarter of the Kelly bet for each pick's score
  • Capped by the bankroll and the pool's depth
RISK RAILS

Capital protection outside the brain

  • Kill switch on drawdown, circuit breaker on failed transactions
  • Every swap is simulated and capped before it is signed
02 · Results

Watch it trade

Panels from the fly's operator console, captured on 2026-10-18.

Live · Robinhood Chain
+6.88 %

Average per trade after costs

Selector backtest
+8.87 %

Per trade over 2,933 held-out trades

Untouched holdout
+5.19 %

Per trade over 2,156 trades

Fly replay
+2.85 %

Per trade over 1,785 trades

Console Trades page for the live Robinhood Chain book: realized P&L +0.0077 ETH, average return +6.88 %, 64 % winners, and a cumulative realized P&L chart.
Live money · Robinhood Chain

The live Robinhood Chain book, trading real ETH: +6.88 % per trade after fees and price impact, 64 % winners.

Console Model page for selector #154: 2,933 test trades averaging +8.87 % against −4.57 % for random picks, profit factor 1.68, with selection, evaluation and holdout periods all profitable.
Backtest · the fly's teacher

+8.87 % per trade after costs over 2,933 held-out trades, against −4.57 % for random picks. On weeks no model ever saw: +5.19 % over 2,156 trades.

Console replay panel: replay passed on Solana and Robinhood Chain; 1,785 replay trades averaging +2.85 % against −4.25 % for random picks, 12 of 16 days profitable.
Replay · the fly's gate

The fly relived 16 days it had never been tuned on, learning as it went: +2.85 % per trade over 1,785 trades, profitable on 12 of 16 days.

Console brain panel: a 3D point cloud of the fly's 41,756 neurons coloured by population, with lines marking the Kenyon cell to MBON synapses it has changed most.
The brain, live

The 41,756 neurons the fly trades with, lit by each minute's scoring, and the synapses its learning has changed most. Play back the last 24 hours, or click a neuron to follow it.

03 · Token

$FLY

The project's token on Robinhood Chain, launched on ponsfamily.com. It exists to gather the community around the experiment. It has no utility today.

Ticker
$FLY
Total supply
1,000,000,000
Chain
Robinhood Chain
Launched on
ponsfamily.com
Contract address
TBA

The contract address above and in the X community is the only one. Anything elsewhere is not us.

Utility

Coming soon. Ties between $FLY and the fly-trader software are forthcoming and will be announced on this page and in the X community. Until they are announced, the token does nothing and confers nothing.

Disclaimer

$FLY is a memecoin. It is not an offer or solicitation to buy securities, and nothing on this site is investment, financial, legal, or tax advice. There is no expectation of profit from the efforts of others; the token may lose all of its value. It confers no ownership, revenue, governance, or other rights, and no utility is promised. Do your own research, understand the risks of ponsfamily.com and Robinhood Chain tokens, and comply with the law where you live. Only spend what you can afford to lose.

04 · Get Started

Run your own fly

Open source under the MIT license. It runs on a Mac, on a Linux PC with an NVIDIA or AMD GPU, or on any 64-bit CPU.

Where the fly runs

BackendHardwareInstall
Apple Metal (MPS)Apple Silicon Mac, macOS 14 or neweruv sync
NVIDIA CUDALinux, NVIDIA GPUuv sync --extra cu126 (or cu130); Docker: docker-compose.cuda.yml
AMD ROCmLinux x86-64: Radeon RX 6000 to 9000, Ryzen AI, Instinctuv sync --extra rocm72; Docker: docker-compose.rocm.yml
CPUAny 64-bit CPUuv sync --extra cpu; the default Docker image

Memory and disk

WorkloadRAMGPUDisk
Trading, Solana16 GB minimumNot neededAbout 2 GB
Trading, + Robinhood Chain16 GB minimumNot neededAbout 40 GB more
Training from scratch64 GB minimumOptional, 4 GB or moreAbout 0.5 TB

You will need

  • Docker, or Python 3.13, uv and PostgreSQLDocker runs everything and keeps itself up to date.
  • Jupiter and Helius API keysHelius's free tier is enough.
  • An Envio HyperSync tokenFor Robinhood Chain. Personal accounts get one free.

Install with Docker: trade with the shipped fly

  1. Download the release and add your keys

    git clone -b distribution https://github.com/bryceweiner/fly-trader.git && cd fly-trader
    cp .env.example .env    # paste HELIUS_API_KEY and JUPITER_API_KEY
  2. Start it

    docker compose up -d --build
    # NVIDIA:  docker compose -f docker-compose.yml -f docker-compose.cuda.yml up -d --build
    # AMD:     docker compose -f docker-compose.yml -f docker-compose.rocm.yml up -d --build
  3. Create and fund the wallet

    Back up .env: it holds the only copy of the wallet's key. Send the wallet 0.5 SOL, then restart.

    docker compose exec fly fly-trader wallet new
    docker compose restart fly
  4. Watch it

    Open http://localhost:8501. It trades paper until 20 trades have closed, then the wallet.

Install from source: train your own

  1. Clone and install

    git clone https://github.com/bryceweiner/fly-trader.git && cd fly-trader
    uv sync                  # macOS: Metal
    uv sync --extra cu126    # Linux, one of: cpu · cu126 · cu130 (NVIDIA) · rocm72 (AMD)
  2. Configure and create the database

    Paste your keys into .env.

    cp .env.example .env
    createdb fly_trader
    uv run fly-trader init-db
  3. Build the brain

    mkdir -p data/brain/raw && cd data/brain/raw
    F=https://raw.githubusercontent.com/snedea/flybrain/main/data
    curl -LO $F/connections.csv.gz -LO $F/neurons.csv.gz -LO $F/classification.csv.gz
    curl -LO https://raw.githubusercontent.com/flyconnectome/flywire_annotations/main/supplemental_files/Supplemental_file1_neuron_annotations.tsv
    cd - && uv run fly-trader build-connectome
  4. Open the console

    Starts every worker and serves the dashboard on http://localhost:8501.

    uv run fly-trader ui
  5. Train

    uv run fly-trader train-selector
    uv run fly-trader train-fly-selector
    uv run fly-trader fly-replay
  6. Go live

    Fund the bot wallet, run the smoke test, then set LIVE_ENABLED=1 (RH_LIVE_ENABLED=1 for Robinhood Chain).

    uv run fly-trader wallet new
    uv run fly-trader swap-smoke --sol 0.01
    uv run fly-trader rh-smoke --yes