Architecture · Engine: Anthropic Claude Opus 4.8
How it works
A scanner-buy-monitor-exit loop. Five components, each an independent daemon on a dedicated VPS, restart-resilient and offline-tolerant. Anthropic Claude Opus 4.8 is the decision engine at the centre — every buy, every exit, every self-grade after a sell runs through it. Below is the user-facing description of each stage. Internal mechanics, weights, and source attribution are proprietary.
How it works
Sweeps fresh graduates and early-stage tokens. Multi-source on-chain intelligence picks candidates.
Claude Opus 4.8 evaluates every candidate across structural signals, holder distribution, demand pressure, and safety. No single signal passes alone.
Live price tracking, intra-trade drawdown awareness, partial-sell ladder for profit-locking.
Profit-lock targets, hard floors on drawdown, structural breakdown detection, peak-and-fade exits — all gated through Claude Opus 4.8.
Claude Opus 4.8 reviews every exit against actual post-sell price action. Pattern of mistakes feeds back into future decisions.
Bot anatomy
Five components, one loop. Each runs as an independent daemon on a dedicated VPS — restart-resilient, no cloud dependency. Claude Opus 4.8 sits at the centre, evaluating every buy/sell decision and grading its own exits.
What the bot is not
- — Not a copy-trade aggregator. We aren’t mirroring whales; the bot makes its own decisions.
- — Not a price-prediction model. There is no neural net guessing the next 5-minute candle.
- — Not a sniper. We do not race graduation transactions; we evaluate after a pool exists.
- — Not custodial. Your funds never leave your wallet.