Skip to main content

TradeFlow

A small, layered, broker-agnostic algorithmic-trading engine. It ships with an Alpaca adapter, but everything above the broker layer is vendor-neutral. It scans a universe of symbols, runs a strategy over them, and either backtests on history or trades live (paper by default) — with optional parameter optimization, walk-forward validation, and constraint-solver portfolio allocation.

Beating the market is hard — embarrassingly hard. TradeFlow's real value isn't a money printer; it's a rigorous skeptic that makes it harder to mistake luck for skill. If your strategy survives walk-forward and the deflated Sharpe, maybe you've got something. If it doesn't, you just saved yourself some tuition.

It is built to be easy to try and easy to read:

  • No TA-Lib, no native build step. Indicators are pure pandas/numpy, so uv sync is the whole install and the Docker image carries no compiler.
  • Broker-agnostic. Every layer is written against a Broker / MarketDataProvider interface; Alpaca is just the first implementation.
  • Strict separation of concerns. Each package does exactly one job.

Two doc tracks:

  • Usage — install, configure, and run the workflows (scan, backtest, live, optimize, walk-forward) plus portfolio allocation and the optional AI agents.
  • Engineering Wiki — how it's built and why, and how to extend it with a new strategy, scanner, or broker.
Educational software

This project is for learning. Trading carries real financial risk. Keep PAPER_TRADE = True unless you fully understand the consequences. No warranty.

60-second tour​

cp config_example.py config.py    # add your Alpaca paper keys
make install                      # uv sync
make scan                         # what's flagged right now?
make backtest                     # scan -> strategy -> performance report

See Installation to get set up.