trade.

synapse trades on uniswap v4 with the brain hook attached, so the LP fee is set per swap by the model and a 0.5% surcharge on the ETH side of every trade is routed to the vault. holders earn from that pool, paid in native ETH, and the claim page handles the rest.

current LP fee-
surcharge-
predicted toxicity-
last swap-

open the pool on uniswap

uniswap routes through the hook automatically. fees you'll pay shown live above; they update on every block as the brain re-scores recent flow.

open uniswap with synapse pool

fee preview for a hypothetical swap

enter an ETH amount and direction. the on-chain model is asked what fee it would quote right now for that exact swap.

predicted LP fee-
surcharge to vault-
total cost on this swap-

how the fee is set

on every beforeSwap the hook reads four features from recent state, calls brain.feeFor(...), and returns the result with the V4 OVERRIDE_FEE_FLAG. uniswap honors the override and uses that fee for the swap, with the LP receiving the full LP-fee amount. the 0.5% surcharge is taken on top from the ETH side via poolManager.take(...) and forwarded to the vault, where it gets unwrapped on its way back out to holders and trainers.

the model can never quote outside the configured bounds, which sit at a 0.05% floor and a 3% ceiling. even if a malicious training step somehow landed (which is not possible while the validation gate is in place), the hook would still clamp every prediction inside those limits.