TLDRocket
Sign in

Stonkfly: a fruit-fly brain trading Bitcoin (market-data inputs, buy/sell/hold decisions)

GitHub

A fruit-fly brain can now make Coinbase trades. It’s real order flow, but nobody’s shown it can actually make money.

Based on reporting by GitHub — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

A research repo called Stonkfly wires a simulated fruit-fly brain to Coinbase and lets it decide whether to buy, sell or hold Bitcoin. The hook is obvious: this is not just a toy model watching charts. It can push actual trading actions through Coinbase integration, while still being framed as a system whose profitable learning has not been proven.

The setup turns public Coinbase prices into an RGB chart, then feeds that into a retained MaleCNS v1.0 graph with 166,700 neurons and 25.6 million connections. On top of that sits a fixed readout that produces the trading decision. If the portfolio moves in the right direction, 15 identified PAM11 dopamine cells get stimulated. If it goes the wrong way, two PPL101 aversive dopamine cells do. A candidate memory rule then alters existing KC-to-MBON connections. The repo is pretty clear that these are engineered reinforcement signals, not proof that the fly brain has somehow become a better trader.

That distinction matters. Synapses changing is interesting. A system making the right call on a few trades is interesting too. But neither one is the same as demonstrating profitable learning, and the source says that straight out. This is the sort of project that will tempt people to blur the line between biological inspiration and actual competence. The line is still there.

The code is set up to be run locally in Python 3.11 with a C++17 compiler, on macOS or Linux. The default mode is paper trading against real public BTC-USDC data with a $100 simulated balance, and it writes logs, sensory images and resumable brain state into runs/paper/. Live trading is gated behind your own Coinbase Advanced portfolio, with at most 100 USDC, a portfolio-scoped ECDSA API key with View and Trade but no Transfer, and explicit opt-in. The defaults are conservative: a $10 maximum order including reserved fees, 24 attempts a day, no shorts or leverage, and a $20 drawdown that stops new orders without liquidating holdings.

My take — AI-written commentary, not fact-checked reporting

This is charming and slightly unhinged in the best way: a fruit-fly brain glued to a crypto API, because apparently every era needs its own laboratory absurdity. The important bit is not the dopamine cosplay; it’s that the repo draws a hard line between flashy reinforcement signals and actual trading skill. That restraint is rare, and more useful than another model with a heroic name and a very expensive ego.

Read more about this at: GitHub

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.