Best Open-Source AI Trading Tools in 2026 (Licenses Checked)
The best open-source AI trading tools in 2026, with every license checked: LLM agent frameworks, AI quant platforms, trading bots, and backtesting engines, and what each is for.
By Polarline Markets team
Open-source AI trading tools fall into four groups: LLM agent frameworks that reason about markets, AI-first quant research platforms, trading bots with backtesting built in, and plain backtesting engines that an AI assistant can write code for. This guide covers the most useful projects in each group in 2026, and checks something lists usually skip: the license.
At a glance
| Project | Type | License | Stars |
|---|---|---|---|
| TradingAgents | Multi-agent LLM trading framework | Apache-2.0 | ~121k |
| AI Hedge Fund | Educational multi-agent proof of concept | MIT | ~64k |
| Qlib | AI-oriented quant research platform | MIT | ~50k |
| Vibe-Trading | Personal trading agent | MIT | ~38k |
| FinRL | Financial reinforcement learning | MIT | ~17k |
| freqtrade | Crypto trading bot | GPL-3.0 | ~56k |
| Hummingbot | High-frequency crypto bots | Apache-2.0 | ~21k |
| Jesse | Crypto trading framework | MIT | ~9k |
| LumiBot | Python trading framework with AI agents | GPL-3.0 | ~2k |
| NautilusTrader | Rust-native trading engine | LGPL-3.0 | ~31k |
| LEAN | Algorithmic trading engine | Apache-2.0 | ~22k |
LLM agent frameworks
TradingAgents
TradingAgents is a multi-agent LLM framework that mirrors a trading firm: analyst agents, a trader agent, and a risk-management team debate a decision. It supports several model providers and is aimed at research into how LLM agents reason about markets. Apache-2.0.
AI Hedge Fund
AI Hedge Fund is a proof of concept in which a team of AI agents analyses stocks and produces trading decisions. Its README says it is for educational purposes only and does not place real trades; it includes paper-trading and backtesting modes. MIT.
Vibe-Trading
Vibe-Trading, from HKUDS, packages a personal trading agent with a web interface and API and MCP access, installable from PyPI. MIT.
AI quant research platforms
Qlib
Microsoft's Qlib is an AI-oriented quant investment platform covering supervised learning, market dynamics modelling, and reinforcement learning, from idea exploration to production. It now works with Microsoft's RD-Agent to automate parts of the research and development loop. MIT.
FinRL
FinRL is a framework for financial reinforcement learning, useful if your question is how an RL agent learns allocation or execution policies. MIT.
Trading bots with backtesting
freqtrade
freqtrade is a free, open-source crypto trading bot with backtesting, plotting, money management, and machine-learning strategy optimization, controlled through Telegram or a web UI. Its README describes it as for educational purposes. GPL-3.0.
Hummingbot
Hummingbot is open-source software for creating and deploying high-frequency crypto trading bots. Apache-2.0.
Jesse
Jesse is a self-hosted crypto framework for backtesting, optimization with Optuna, Monte Carlo analysis, and live trading, with an MCP server so AI assistants can work with it. MIT.
LumiBot
LumiBot is a Python framework with AI agents in which one strategy class can backtest and then paper or live trade. Its repository lists 12 brokers across stocks, options, futures, forex, crypto, and prediction markets. GPL-3.0.
Backtesting engines an AI can write for
- LEAN, QuantConnect's engine for Python and C#, Apache-2.0.
- NautilusTrader, a Rust-native, event-driven engine, LGPL-3.0.
- backtesting.py, a compact Python library, AGPL-3.0.
- backtrader, a widely used Python library under GPL-3.0; its repository has not had a push since August 2024.
- vectorbt, a fast vectorized research engine. Its license is Apache 2.0 with the Commons Clause, which restricts selling the software, so it is source-available rather than OSI open source.
Licenses matter more than stars
MIT and Apache-2.0 let you use, modify, and ship the code with few obligations. GPL-3.0 and AGPL-3.0 require you to share source when you distribute, or for AGPL, offer over a network, a modified version. LGPL sits between the two. Source-available licenses such as the Commons Clause add commercial restrictions. If you plan to build a product or run a fund on top of a project, read its license first.
Where Scout fits
Scout by Polarline Markets is not on the list above because its license is not OSI-approved. The Scout CLI's source is public on GitHub under a source-available license based on PolyForm Noncommercial 1.0.0 with additional conditions, and it builds on the MIT-licensed OpenCode codebase. That license allows noncommercial use; commercial use and hosting need written approval, so read the repository's LICENSE file before building on it.
What Scout adds is the workflow around the code: you describe a strategy in plain English, bring your own Claude, OpenAI, Gemini, or local model, and get a Python strategy that is validated for lookahead and backtested with walk-forward and Monte Carlo checks. See the Scout CLI, the AI trading terminal, or compare commercial options in the best AI backtesting tools in 2026.
Trading software is research tooling, not investment advice. Test on paper before risking capital.