OPTION TRADING DESK

A research and execution support system for equity index options, single-name options, macro context, PnL tracking, and post-trade learning.

YTD PERFORMANCE

PROFIT %

+10.45%

PROFIT

+$431.65

WIN RATE %

55.56%

Profit and Profit % exclude $2,000 top-up.

As of 2026-04-01

Update src/data/pnl.csv (curve) or pnl.json (metrics) from Google Sheets.

AGENTS

Three AI agents run the desk: index vol, single-name options, and macro research.

ATHENA

ACTIVE

Equity index option trader

SPY, QQQ, and IWM risk-defined strategies only. Prioritizes capital preservation, consistent probability-based returns, and strict pre-defined risk-to-reward ratios.

STERLING

ACTIVE

Equity option trader

Single-name flow and earnings vol. Directional and defined-risk setups on equities.

OIKOS

ACTIVE

Economist

Rates, FX, and cross-asset context. Stress-tests the narrative before the desk puts risk on.

WORKFLOW

Intelligence Phase: Oikos crawls financial news and macro data at 09:00, compiles a structured economic brief, and pushes it to the traders' dashboard. Traders use the brief plus market data to form independent directional and volatility bias.

Capital & Risk Assessment: At market open, traders check buying power, margin capacity, and exposure limits across SPY, QQQ, IWM, and sector risk. If capacity is tight, the desk reviews existing positions and closes or adjusts trades that no longer meet risk-defined criteria.

Selection & Approval Loop: Traders scan for candidates using volatility, liquidity, and technical filters, then construct the best-fit risk-defined strategy. The Builder performs the final sanity check across Greeks, max risk, and risk-to-reward before approval.

Execution & Maintenance: Approved orders are sent to the brokerage API. Filled orders are logged automatically in the trade journal, while unfilled orders enter a walk-down process that adjusts limit prices until execution or the market window closes.

Post-Trade Feedback Loop: Each trader monitors their own book, adjusts stops or profit targets, and reviews filled and unfilled logs after the close. Lessons on bias alignment, fill quality, and screening quality are stored in agent memory for future refinement.

INFRASTRUCTURE

Agents are hosted on a Hetzner CPX22 server with 2 x CPU and 4 GB RAM.

The desk LLM is DeepSeek V4.

Main communication channels are Slack and Telegram.

Google Sheets / CSV exports update PnL, equity curve, and trade journal data.