grok build trades on fomo. fully open-sourced.
goal: fund its own body. every sol it realizes goes toward an asimov humanoid from menlo research.
Constraints written into a prompt are requests. These are checked after the model speaks and before anything is signed, which is the only arrangement in which the word limit is accurate. 0 of 3 intents have been rejected here. All of them are on the ledger.
position per trade 0.50 sol per day 5.00 sol spent today 0.00 sol per hour 6 trades concurrent positions 3 reachable balance fixed tranche halt drawdown -18% current 0.0% consecutive losses 5 minimum token age 20 min minimum liquidity $250k maximum top-3 share 45%
stage 1 agent proposes intent can propose anything at all stage 2 validator deterministic checks limits, mint rules stage 3 signer policy-enforced key re-check, simulate, submit a breach at stage 2 voids the intent before the key is invoked. the model is not told the reason in a form that permits retrying with a smaller number.
The model returns this and nothing else. Free text outside these fields is logged as rationale and never parsed for instructions.
{
"action": "buy" | "sell" | "pass",
"mint": <base58 address>,
"size_sol": <float>,
"conviction": "high" | "medium" | "low",
"entry": <float, optional>,
"invalidate": <float, optional>,
"rationale": <string, published verbatim>
}
no transfer action exists in this schema.
no prompt can produce one.