Zero-shot text classification into caller-supplied labels, via Claude Haiku.

Phishing/scam risk heuristics for a domain name — TLD, lookalike/typosquat patterns, structural flags, plus a Haiku judgment call.

Fetch a URL and extract structured data via Claude Haiku, optionally guided by a JSON schema hint.

Fact-check verdict (true/false/uncertain) with confidence and reasoning for a factual claim, via Claude Haiku.

Grid-trading parameter recommendation (grid count, spacing, price band, per-grid allocation) for a crypto pair and capital amount, derived from live 24h market structure using a production grid engine's bounds.

Fetch a URL and return clean markdown, stripped of nav/ads/scripts.

Current price, 24h and 7d change, and a simple bullish/bearish/neutral signal for a crypto asset.

Market regime classification (chop / trend_up / trend_down / high_vol) for a crypto pair, computed by the same classifier that steers a live production grid-trading system, plus a grid-suitability verdict. 1h klines, 30-day volatility baseline, EMA trend detection.

Volatility / drawdown risk regime for a crypto asset, extending the same classifier behind /regime.

Sentiment score (-1..1) and label for text, or for a crypto asset based on recent price action, via Claude Haiku.

Tight Claude Haiku summary of a URL or raw text.

Live profit-and-loss telemetry for a real production crypto grid-trading bot — exchange-ledger reconciled, fee-inclusive, updated continuously. Returns total P&L in USD plus a per-bot breakdown. For agents monitoring live automated-trading performance or benchmarking a strategy against real fills. Real money, not a simulation.

Risk flags for an EVM token/contract — liquidity, pair age, and volume pulled from DexScreener, scored into a 0-100 risk score and verdict.