AI tools were used to assist with parts of the code, documentation and graphics of this item. The author directed the work and reviewed the code before release.

What it does

• Paper, alert or live mode: live trades pump.fun with your keys (Hyperliquid with the optional perps module), capped, dry run first
• AI analyst: Claude briefing every few hours (market read, risks, stance per book, watch list)
• pump.fun real-time agent with EV gate, fractional Kelly and bonding-curve fills
• Four extra pump.fun lanes: moonshot, graduation, revival, smart-money consensus
• Exit-quality grades after every pump.fun close
• Attested sha256 journal with `verify`
• Per-trade cost accounting (fee, slippage, tip, failed tx) in the journal and report
• Wallet book from the on-chain stream: realised PnL, bot/sniper/dust/creator/listed-insider flags, out-of-sample persistence test

You need

• Node.js >= 22 (package.json engines ">=22"; global WebSocket and process.loadEnvFile; src/lib/doctor.mjs). Docker image
• `npm ci` then `npm run build` (tsc compiles the operator console in src/admin and packages/admin-kit to dist/; typescrip
• One long-running process on a machine with a disk: VPS/container via docker compose, or PM2 (README.md:62-69). Back up d
• ADMIN_SECRET_KEY recommended on a server (.env.example: "Required on a server"); otherwise generated once into data/.adm
• Recommended: a private Solana RPC/WebSocket (Helius, QuickNode, Chainstack), set as a console secret (src/admin/index.ts