FruitFly Capital
We gave a swarm of fruit-fly brains control of a crypto fund.
Demo
About
What happens when you let fruit-fly brains manage a crypto fund? Fruit Fly Capital turns the market into a living ecosystem and lets a swarm of simulated flies explore it. Every token becomes a place in the world, shaped by real market conditions. The flies decide where to gather, when to stay, and when to move on. Their collective behavior becomes the fund’s portfolio, while a separate risk layer keeps positions within safe limits. No human trader is picking the assets, and no LLM is choosing what to buy. The market becomes the environment, the swarm reacts to it, and capital follows.
How it's made
Fruit Fly Capital combines live onchain market data, a biological nervous-system simulation, a 3D world, and DeFi infrastructure. The Graph is used to pull deeper onchain market and pool activity, while DexScreener and CoinMarketCap help discover and enrich the token universe. We transform those financial signals into a physical environment: liquidity, activity, flow and risk become things the flies can sense such as odor, light, motion and environmental intensity. The brains are based on MaleCNS v1.0, the fruit-fly central nervous system connectome from Janelia/Google Research. We use Brian2 in Python to run persistent neural simulations for 16 independent flies. Each fly has its own neural state and receives only sensory information from the world; it never receives a token name, price target or direct “go buy this” command. For the body and flight representation we use the open-source TuragaLab FlyBody Drosophila model, with a browser-side embodied simulation and neural-to-motor interface. The visual world is built in TypeScript with Vite and Three.js, including the city, token habitats, fly bodies, lighting, post-processing and realtime movement. The frontend talks over WebSockets to the Python brain server, so every sensory update can flow through MaleCNS and come back as a motor command. The public demo renders 80 flies, with 16 independent CNS agents and additional follower bodies for scale. On the DeFi side, we built a Solidity / Foundry fund vault using OpenZeppelin contracts and deployed it on Robinhood Chain testnet. Deposits are accounted in WETH and mint FFC fund shares. Privy provides the server-side wallet and transaction-policy boundary for the strategy treasury, and the execution layer is designed to route approved trade intents toward Uniswap. The important separation is that the flies decide the raw allocation, while normal risk controls decide whether and how much can actually be executed. The hacky part is tying all of these layers together in realtime: live markets → sensory environment → biological neural activity → physical fly movement → swarm allocation → guarded DeFi execution. Instead of training a model to predict prices, we turned the market into an ecosystem and let a swarm of simulated nervous systems explore it.