CryptonBot Pro+
Async crypto trading bot with ML signals and risk management
I've been building CryptonBot Pro+ for about a year as a solo side project. Today I'm opening a closed beta.
The core problem I kept seeing: most trading bots fail not because of bad strategies, but because they have no defined failure modes. What happens when the exchange goes down mid-trade? When your drawdown hits 20%? When an order gets stuck?
So I built the risk management layer first.
What's inside:
- Async Python (FastAPI + ccxt), Binance Spot
- Grid & Adaptive Grid strategies (BTC, ETH)
- ML signals: DQN + Ensemble (PyTorch)
- Drawdown guards, position sizing, idempotent orders
- 238 tests, CI/CD, Prometheus + Grafana
Current status: testnet verified, looking for 10 beta testers.
Happy to answer any questions about the architecture or approach.
https://gumroad.com/l/wpvcv
1 Comment
This matches a pattern I keep finding when I audit other bots' code: it's rarely the strategy that blows up an account, it's what happens when the exchange goes quiet mid-trade. A recurring bug I've seen in the wild — get_open_orders() (or the equivalent call) silently returns an empty list on any API exception, and the reconciliation loop reads "no orders returned" as "all orders filled," so a live position gets marked closed while it's still open on the exchange. Curious how CryptonBot's risk layer treats an ambiguous or failed exchange response for open positions specifically — retry-with-backoff, halt-and-alert, or something else? That edge case is usually harder to get right than the drawdown/position-sizing math itself.