Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
// Oracle — outcome grading (pure).
//
// Closes the learning loop. Once the data brain labels a coin's ground-truth
// outcome (graduated / rugged / all-time-high multiple), every agent action on
// that coin is graded here: did the conviction call pay off? This is what turns
// the action ledger into an honest win-rate record and what lets the conviction
// backtest report real numbers instead of promises.
//
// Pure: takes one action row + the coin's outcome and returns the settled
// fields. The worker does the I/O.
const num = (v) => (v == null ? null : Number(v));
/**
* @param {object} action oracle_watch_actions row { size_sol, entry_mc_usd }
* @param {object|null} outcome pump_coin_outcomes row { graduated, rugged, ath_multiple, last_market_cap_usd }
* @returns {{ settled:boolean, outcome:string, peak_multiple:number|null, realized_pnl_sol:number|null }}
*/
export function gradeAction(action, outcome) {
if (!outcome) return { settled: false, outcome: 'open', peak_multiple: null, realized_pnl_sol: null };
const graduated = !!outcome.graduated;
const rugged = !!outcome.rugged;
const peak = num(outcome.ath_multiple);
const entryMc = num(action?.entry_mc_usd);
const lastMc = num(outcome.last_market_cap_usd);
// Nothing to grade yet — outcome row exists but carries no resolved signal.
if (!graduated && !rugged && peak == null && lastMc == null) {
return { settled: false, outcome: 'open', peak_multiple: peak, realized_pnl_sol: null };
}
const size = num(action?.size_sol) || 0;
// Honest mark-to-market: value vs. entry from the last observed market cap.
// We don't model an exit, so this is the position's standing mark, not a
// realized sell. Null when we can't anchor an entry.
const mark = entryMc && lastMc ? lastMc / entryMc : null;
const realized_pnl_sol = mark != null ? +(size * (mark - 1)).toFixed(6) : null;
// Loss conditions outrank a peak-based win: a coin that spiked 2× and then
// rugged (or is marked down >50% from entry) was not a winning call — nobody
// systematically exits at the wick. Graduation alone overrides, since a
// graduated coin has a real market regardless of the later pump.fun mark.
let label;
if (graduated) label = 'win';
else if (rugged || (mark != null && mark < 0.5) || (peak != null && peak < 1.2)) label = 'loss';
else if (peak != null && peak >= 2) label = 'win';
else label = 'flat';
return { settled: true, outcome: label, peak_multiple: peak, realized_pnl_sol };
}
/**
* Roll a set of graded actions into an agent's win-rate summary. Pure.
* @param {Array<{outcome:string, realized_pnl_sol:number|null, size_sol:number}>} actions
*/
export function summarizeActions(actions = []) {
let wins = 0, losses = 0, open = 0, pnl = 0, sized = 0;
for (const a of actions) {
if (a.outcome === 'win') wins += 1;
else if (a.outcome === 'loss') losses += 1;
else if (a.outcome === 'open' || a.outcome == null) open += 1;
if (a.realized_pnl_sol != null) pnl += Number(a.realized_pnl_sol);
if (a.size_sol != null) sized += Number(a.size_sol);
}
const resolved = wins + losses;
return {
total: actions.length,
wins, losses, open,
win_rate: resolved ? Math.round((wins / resolved) * 100) : null,
realized_pnl_sol: +pnl.toFixed(6),
deployed_sol: +sized.toFixed(6),
roi_pct: sized > 0 ? Math.round((pnl / sized) * 100) : null,
};
}