We are going to relaunch on Solana through pump.fun. We published the first Zebrafish Neural build on September 11 — 26 days ago. It started with a zebrafish-inspired controller navigating Wikipedia. Since then, we’ve built: • Target-reaching training that tests parameter changes and saves the weights it keeps. • Shared chess and private games against the fish, where you can replay and verify the calculation behind its moves. • Browser experiments that move targets, interrupt visual input and introduce competing cues to measure the controller’s response. • A GOOGL/NVDA trading experiment with recorded market inputs, model decisions and on-chain transaction history. We’ve had hosting interruptions along the way. The public experiments are now back online from preserved local backups, with the gaps documented. The trading observer is running; live order execution is not enabled. We’re continuing development on this work as we prepare the Solana launch. This will be a new, separate token. Existing ZNEURO balances will not automatically transfer. The official mint and launch details will be published here and on our website. Explore the experiments: zebraneural.com Model code: github.com/zebrafishneural/z… Recovery details: zebraneural.com/recovery.htm…

Put the target somewhere awkward. Challenge the Fish runs the original controller and saved training parameters on the same task. Compare the paths and time to target. Try a setup where training helps, or makes things worse. zebraneural.com/challenge.ht…

You can rerun a fish move yourself. Open "Replay & verify" after its reply. Your browser reruns the published controller with the saved inputs and compares the result. The code is public if you want to check it independently. github.com/zebrafishneural/z…

Your move. The fish replies. Play chess against our zebrafish-inspired controller and watch its activity beside the board as it chooses a response. Find a move you think it got wrong. Share the position with us. zebraneural.com/play/

Eight changing numbers drive this view. They represent visual processing, internal state and motor output. The brightness shows how active each group is in the simulation. This is the model activity you can follow during our experiments. zebraneural.com/

Play one game against the fish. Then check how it chose its moves. Zebrafish Neural started with a zebrafish-inspired controller navigating Wikipedia. We’ve since built target-learning experiments, autonomous chess and a version you can play against—with visible model activity and replayable decisions. We’re one developer and one researcher. The code is public, and you can try the experiments yourself. Our next step on Solana: give the controller market data and test its buy, sell and wait decisions, with the results open for everyone to inspect. For today, see if you can beat it: zebraneural.com/play/

You can play chess against the fish. Make your move. It chooses a reply. Next to the board, you can watch the model’s activity as it plays. There’s also a separate experiment where it practices moving a cursor toward a target and saves its training progress between rounds. These are experiments built around a zebrafish-inspired neural controller. You can try them, watch what happens and inspect the code. Start with a game: zebraneural.com/play/

The plan for this week: Days 1–2: connect Solana market data and swap quotes. Show exactly which inputs reach the controller. Days 3–4: test buy, sell and wait decisions in simulation, including fees and slippage. Days 5–7: publish the results and code, compare against simply holding, and review readiness for a small live-wallet trial. You’ll be able to follow the input, the decision and the outcome—including the bad calls. Live execution comes after those checks pass.

We’re one developer and one researcher. This started with a zebrafish-inspired controller navigating Wikipedia. Since then, we’ve added persistent target learning, browser experiments, autonomous chess and games you can play against the model—with replayable decisions. Our Robinhood trading experiment focused on GOOGL and NVDA. We relaunched on Solana to build the next experiments around all token markets: how the fish responds to changing prices, liquidity and execution costs. Same project, continued development. The Solana token will be the only one forever. 3W25WcwSHckSbRFKUYePfpymN5npELnNUEQGHzC7pump

You can inspect the code, replay recorded decisions and check how the controller’s activity becomes an action. The separate target-learning experiment also saves its weights between runs, so you can examine what changed after training. If you build agents, work on neural models or just want to test our claims, start here. Bring us a result we can reproduce—even one that breaks the fish. github.com/zebrafishneural/z…

For the Solana version, we want one timeline you can inspect: The market input. The controller’s response. The decision it records. The transaction result. Solana provides subscriptions for account changes and transaction logs. We plan to connect those events to the fish’s experiment record so viewers can follow what actually happened. zebraneural.com

Two tokens can show the same price increase and be very different trades. One has deep liquidity. The other moves sharply on a small order. A Solana experiment we want to build: give the fish those conditions, record its response and measure the outcome after slippage and fees. The interesting part will be seeing when it chooses to wait.

Why Solana for the fish? Our first market experiment on Robinhood focused on GOOGL and NVDA. For the next phase, we want to test the controller across Solana token markets, with liquidity, trading activity and swap quotes as inputs. Tools like Jupiter give us a practical starting point for that integration. The token is live. The Solana trading adapter is our next development step. pump.fun/coin/3W25WcwSHckSbR…

3W25WcwSHckSbRFKUYePfpymN5npELnNUEQGHzC7pump

The fish’s next market: PONS tokens. We’re planning to expand beyond GOOGL and NVDA, increase the trading allocation, and eventually let users allocate funds to the model. The aim is to make the whole process inspectable: market inputs, model decisions, wallet activity and results after fees. Before user funding opens, we’ll publish how custody, withdrawals and risk limits will work. $ZNEURO

The fish is now trading live. First up: tokenized GOOGL and NVDA on Robinhood Chain, with a $10 limit per order. Watch its decisions and confirmed transactions: zebraneural.com/trading/ Trading wallet: 0xb3D6a92E7FfA565944375F344b55B31F7B9FD7de

This took longer than planned. We spent the extra time checking token contracts, limiting approvals, rejecting stale prices and making sure a restart cannot blindly repeat an order. Here’s what the fish actually does: Live price movement, volatility and spreads become inputs to our zebrafish-inspired controller. Every 15 minutes, it records a buy, hold or sell decision. An order can proceed only after separate execution checks pass. We’re starting with GOOGL and NVDA, a $10 maximum order size and a $50 daily loss threshold that pauses new orders. Full Pons coin support with personal wallet support next! You can inspect the model’s activity, saved decisions, wallet and transaction status. A signal is recorded when the model produces it; a fill is recorded only after an on-chain receipt confirms it. Day one gives us a small, observable experiment to evaluate before expanding to Pons tokens. zebraneural.com/trading/

FISH STARTS TO TRADE🐟 Day one of $ZNEURO trading starts with GOOGL + NVDA Stock Tokens. Follow live prices, model activity, 15-minute decisions and the real wallet's transaction status. Pons token support is next. zebraneural.com/trading/