Beleth is an autonomous trading agent that sells option premium, exploiting the gap between implied and realized volatility (the volatility risk premium), using exclusively defined-risk structures, multi-leg short vertical credit spreads submitted as a single order, never naked legs.
Key points of the system:
- Strategy: short vertical credit spreads on elevated implied volatility.
- Decision engine: an LLM (via LiteLLM, to stay provider-agnostic) evaluates each cycle against the evidence package — spreads, VIX, term structure, macro events — and proposes or rejects the trade.
- Multi-layer risk system, fully live in production: minimum VRP threshold to open, gates on inverted term structure and macro events, automatic exit at +50% profit / -2x loss, 1-2% sizing per trade, 3% daily drawdown stop, size taper based on the historical VIX percentile (1990-2026), anti-stacking block on resting orders, aggregate cap on committed risk.
- Radical transparency: the public dashboard shows every cycle, every executed trade, and — most importantly — every rejected trade, with the full reasoning. That's not a side detail, it's the heart of the project. A bot that knows how to say no is more interesting than one that fires off trades constantly.
- Deployment: Docker container on dedicated hardware, Postgres (Supabase) as the single source of truth, Next.js dashboard on Vercel.
Live dashboard: https://beleth.davidemaiorana.dev/dashboard - Public repo: github.com/Akaired/beleth-agent
The character I wanted to give it: profitable, cautious, winning. Not the most aggressive bot possible, the most rigorously disciplined one.