OSS version of original aequify rewritten in Python + Mojo.
- Binance Futures USD-M
APEX is a GPU-accelerated parameter optimization engine for trading signals. It uses Mojo kernels to perform high-speed grid search over parameter combinations.
In other words, just a fancy backtester... at least for now.
How it works:
-
Bootstrap - Analyzes historical trade data to find optimal signal parameters via GPU grid search
- Computes rolling high/low prices over configurable time windows
- Calculates volume delta (buy/sell imbalance) over the same windows
- Tests thousands of parameter combinations in parallel on GPU
- Selects best parameters based on win rate and PnL
-
Live Detection - Real-time signal detection using bootstrapped parameters
- Tracks rolling metrics from live trade stream
- Emits LONG signals on price drops with selling pressure
- Emits SHORT signals on price rises with buying pressure
Configuration is managed via config/config.yaml. Key sections:
trading- Trading mode, position sizing, risk limitsfilters- Symbol filtering (whitelist/blacklist, price change filters)engines.apex.bootstrap- APEX parameter bounds and grid search settingsdatabase- QuestDB connection settings
Per-symbol overrides can be placed in config/overrides/{SYMBOL}.yaml to customize parameter bounds for specific trading pairs. These are auto-generated after bootstrap with narrowed bounds based on good-performing combinations.
- mise - Runtime manager
- uv - Python package manager (installed via mise)
- Python 3.14+
- QuestDB (auto-installed via mise tasks)
# Install dependencies
mise install
mise run install
# Run with dev dependencies
mise run dev
# Run aequify
mise run aequifyRun mise run to see all available tasks.
GPU: NVIDIA GPUs supported and tested. AMD may work but untested. APEX bootstrap requires GPU for grid search acceleration.
Join the Discord: https://discord.gg/RcMGpasGC7
Not accepting PRs at this time. Issues and feature requests welcome.
If you find this project useful, buy me a cigarette
