Axonex

Trade 120+ perpetual markets with oracle-backed pricing, $AXON Ca : 4hfVnxzUajXPgTajoDM63v3zBP9WA4yKGi8829YHpump

paying dex..

the github! github.com/Axonexdev/Axon…

reflection is the part that compounds. when a trade resolves, axonex scores it — win / loss / breakeven — and writes the completed episode back into memory. so the history that informs the next decision is always grounded in what actually happened, not what the model hoped would happen.

the core trick is a regime fingerprint. every market moment gets compressed into a signature: trend bucket · volatility bucket · funding sign · OI skew · basis when a new setup appears, the agent fingerprints it and pulls the most similar PAST setups — but only ones that already resolved, so each recalled memory carries a real win or loss

axonex runs a loop, not a prompt: perceive — read the market recall — pull its own past trades in similar conditions reason — decide with that track record in context act — open the position reflect — score the outcome, write it back to memory every cycle makes the next one smarter

most "AI trading agents" are a goldfish with an API key. feed the model the current price → get a signal → forget everything. no memory of yesterday's call. no record of whether it was right. no idea this exact setup has burned it three times before. axonex is built the opposite way