🤓 Latest Submissions
Investonaut
Investonaut is an autonomous AI trading agent — not a script — that runs a bull / bear / neutral debate through an LLM risk arbiter, reads live Alpaca option chains via the Alpaca MCP server, picks a Level-3-legal options structure by IV rank, and sizes every idea at ≤2% equity behind a hard suite of risk gates. It harvests theta when implied volatility is rich and buys cheap protection when it's low — and a walk-forward out-of-sample gate decides what gets to trade at all. *Bull sub-agent* proposes covered calls — sell an OTM call (~Δ0.25, ~21 DTE) against shares it already holds, harvesting theta + volatility premium when implied vol is rich. *Bear sub-agent* proposes a protective put / credit collar — buy an OTM put on SPY to cap drawdown on the long book, funded by selling an OTM call → a near-zero-cost hedge when implied vol is cheap. *Neutral/Income sub-agent* proposes a cash-secured put — sell an OTM put, fully cash-collateralized, to buy dips at a discount while collecting premium. *IV-Rank Scorer* computes each underlying's implied-vol percentile and routes the debate: high IVR → sell premium; low IVR → buy cheap protection. This is the agent's market-condition judgment. *Risk Arbiter* — the referee. Every proposal must clear it before any order is placed. It enforces the risk framework and records rejections as loudly as fills. *LLM Referee* — a local LLM (Ollama, `qwen2.5:1.5b`) reviews every accepted proposal; its verdict and reasoning land in the same SQLite audit log. Advisory by default — the hard gates in code do the enforcing, so the LLM can never silently approve or block a trade. Options express conviction with defined risk and monetize time decay. Investonaut's core thesis: *equities drift, but options harvest the volatility-risk premium.* Every trade carries an explicit, capped risk-reward — never a naked bet. All structures are strictly Alpaca Level-3-paper-legal — no naked shorts, ever.
4 Sep 2026
