
AegisFlow
A regime-aware trading agent for BTC and ETH on OKX TR. Every proposal passes an independent risk governor before it becomes an order.
Links
My role: Solo: everything, from scratch.
Problem
An autonomous trading agent is hard to trust if nobody can see why it trades or why it waits. A bad proposal also has to be stopped before it reaches the exchange, not explained afterwards.
Approach
AegisFlow splits the agent into three processes. The brain ranks momentum and mean-reversion signals for BTC and ETH by their edge after costs, spread and impact, and proposes only the best one. A separate guard owns the money: an independent Risk Governor checks freshness, limits and losses in a fixed order and fails closed.
Result
AegisFlow was built in one day at the Agentic Trading Hackathon in Komünite Space and ran a PAPER session with 30 USDT. The report shows 19 cycles that ended without a signal and no trades. The code, the risk rules and a list of eight known limitations are public on GitHub.
Highlights
- Built in 1 day
- 3% session loss limit
- 0.5% risk per trade
- Fails closed
Stack
- Python
- SQLite
- OKX Agent Trade Kit
- OpenAI
- cloudflared