TorchTrade is a modular reinforcement learning (RL) framework designed to make advanced RL methodologies accessible for algorithmic trading. It targets researchers, engineers, and power users by providing a unified platform for both backtesting and live trading, integrating RL with traditional and LLM-based trading strategies. The primary benefit is enabling rapid deployment and experimentation of sophisticated trading agents.

How It Works

Built upon TorchRL, TorchTrade supports a wide array of RL paradigms including online, offline, and model-based learning. Its core innovation lies in its modular design, offering environments for both live trading across major exchanges and rigorous offline backtesting. The framework uniquely integrates traditional rule-based strategies and cutting-edge LLMs (both local and frontier models) as adaptable trading actors, facilitating a blend of established and novel approaches. This allows for a seamless transition from research to production using consistent codebases.

Quick Start & Requirements

Installation is streamlined using the UV package installer:

1. Install UV: curl -LsSf https://astral.sh/uv/install.sh | sh
2. Clone the repository: git clone https://github.com/TorchTrade/torchtrade.git
3. Navigate and sync dependencies: cd torchtrade && uv sync
4. Activate the virtual environment: source .venv/bin/activate

Optional dependencies for LLM actors or Chronos forecasting can be installed with uv sync --extra llm or uv sync --extra chronos, respectively. Prerequisites include Python 3.8+; CUDA is recommended for GPU acceleration. Official documentation and examples are available on the project's website.

Highlighted Details

• Supports multi-timeframe observations (e.g., 1m, 5m, 15m, 1h) for richer market context.
• Implements multiple RL algorithms including PPO, DQN, IQL, GRPO, DSAC, and CTRL.
• Offers direct live trading integration with major exchanges like Alpaca, Binance, Bitget, and Bybit.
• Features LLM integration (e.g., GPT-4o-mini) and rule-based actors for diverse strategy development.
• Provides pre-trained encoder transforms for time-series foundation model embeddings.

Maintenance & Community

TorchTrade is under active development, with continuous feature additions and improvements. Users should anticipate potential API changes. Support and further information are available via email at torchtradecontact@gmail.com and through the project's comprehensive website and documentation.

Licensing & Compatibility

The project is released under the MIT License, which is permissive and generally suitable for commercial use and integration into closed-source projects.

Limitations & Caveats

As a project under active development, TorchTrade is a "Work in Progress," and users should be prepared for evolving APIs and features. The current scope is limited to single-asset trading environments; multi-asset portfolio optimization and cross-asset trading capabilities are planned for future releases.