Skip to main content
Retail and prosumer traders today face three structural problems with automated trading:

​
1. Unverifiable track records

Most trading bots advertise PnL screenshots, cherry-picked backtests, or dashboards that the operator fully controls. There is no way for a user to independently confirm:
  • That the strategy actually ran with the parameters claimed
  • That the reported outcome matches the real fills
  • That past “winning runs” weren’t selected after the fact
Trust is asked for, not earned.

​
2. Custodial risk

Many bots require users to deposit funds into a platform wallet or hand over withdrawal-enabled API keys. That turns a trading question into a counterparty question: even if the strategy is good, the user can still lose everything to a hack, a rug, or a frozen withdrawal.

​
3. Raw-PnL optimization

Bots optimized purely for absolute PnL tend to:
  • Take uncapped tail risk to chase returns
  • Look great until a single regime change wipes them out
  • Hide drawdowns, leverage, and variance from the user
Risk-adjusted performance — the only thing that compounds — is rarely the optimization target.
Perps Agent is built to address all three: verifiable runs on-chain, non-custodial execution on Bybit, and a risk-adjusted optimization target baked into the AI loop.