Quantis AI Options
A barbell: a heavy, defined-risk core of credit spreads and iron condors that earns the base return, plus a light, hard-capped convex sleeve for asymmetric upside. Bounded downside by construction.
Personal engineering project · Paper trading only · Not investment advice
Most of the book sits in defined-risk short premium — credit spreads and iron condors that win the majority of the time and can only lose a known, capped amount. A small sleeve takes the convex bets that engine can't: long premium, then disciplined naked and 0DTE. The barbell shape — heavy safe end, light risky end, nothing in the fragile middle — is what keeps the downside bounded while leaving room for asymmetric upside. The core proves itself first; the sleeve is earned phase by phase.
Two ends of a barbell
Heavy and defined-risk on one end, light and convex on the other. Position size is the risk control, not conviction.
Core
Put credit spreads and iron condors at ~16Δ, 45 DTE. Close at 50% of max profit, mandatorily manage at 21 DTE, size every position to lose no more than 1% of NAV. Theta-positive and fully defined-risk — this is the heavy, stable end of the barbell that produces the base return.
Sleeve
A small, hard-capped allocation to convex bets the core can't express: directional long premium (Phase 2), then short naked puts / strangles with hard stops (Phase 3), then 0DTE (Phase 4). Capped at 25% of NAV. This is the light end — limited downside, asymmetric upside.
Phased rollout
Each phase has to clear its weekly-review gate before the next risk tier switches on. Shipping the whole toolbox at once is how options books blow up.
Core only
Put credit spreads + iron condors trade alone. Proves the defined-risk engine, multi-leg fills, and the drawdown kill-switch with real data before any convexity is added.
Add directional sleeve
Long calls / debit spreads on the same signals as the core. Defined-risk by construction — the gentlest convexity, switched on first.
Naked, strangles, 0DTE
Short naked / strangles with hard stops and VIX gates (Phase 3), then 0DTE at the smallest sizing (Phase 4). Each unlocks only after the prior phase clears its review.
Hard rules the LLM cannot override
The old “never naked” rule was replaced — not removed — with disciplined controls enforced before any order is placed. The drawdown kill-switch is the backstop under everything.
- −15% drawdown-from-peak kill-switch — sticky, resets only at the Friday weekly review
- Sleeve hard-capped at 25% of NAV (concentration & convexity limit)
- Naked positions carry hard stops: 2× credit (puts), 2.5× (strangles)
- Per-position max loss ≤ 1% NAV on the core; notional caps 5% / 7%
- VIX gates: no naked above 25, halt all new entries above 40
- Buying-power floor 20% (never fully deployed)
- Earnings buffer DTE+5 — hard requirement for any naked position
- Mandatory 21-DTE management on every short option (gamma cliff)
What this can realistically deliver
The infrastructure — an autonomous LLM agent, hard-rule discipline enforced in code, git-as-memory, atomic state — has engineering value regardless of P&L. No signal service, no subscription, nothing for sale, no community.
How it's built
No databases, no ORM. Strategy rulebooks, trade ledger, weekly reviews — all markdown committed to a private GitHub repo. Every routine run is a fresh container that clones, decides, commits, exits. A single atomic JSON state file holds the core/sleeve maps, NAV peak, kill-switch, and VIX regime cache.
See it run
The dashboard is private. The only person with access is the operator. Click below if that's you.
Sign inLooking for the stocks bot? Quantis AI Trading landing →