Top 10 AI Trading Repos on GitHub | Most Stars Isn't the Same as Best

TradingAgents copies the org chart of an investment firm. Four analyst agents study fundamentals, the chart, news sentiment and the economy. A bull agent and a bear agent then argue over what the analysts found.

A trader agent weighs the debate, and a risk manager has to approve before any idea becomes a signal. You choose which AI powers each seat, so a cheap model can read data while a smarter one makes the call.

One AI prompt tends to agree with itself. Forcing two agents to argue, then making a third approve, is a smart answer to that problem, and it is easy to watch it work.

OpenBB pulls market data from over 100 providers into one place, so you stop writing a new connector for every feed. It comes as a Python toolkit, a desktop workspace with charts and dashboards, and a layer that lets AI assistants fetch quotes and filings.

It is a research tool, not a trading bot. It finds setups and cleans data, then hands off to whatever places your orders.

Market data is scattered across dozens of APIs that all speak differently. Connect once and use the data everywhere is a promise every quant wants.

ai-hedge-fund builds an investment committee out of AI characters inspired by famous investors, from Warren Buffett's value lens to Michael Burry's hunt for red flags. Each one judges a stock its own way.

Quantitative agents add the numbers, and the output is a memo, a conviction score and suggested position sizes. It is a teaching project, and the author says so.

It turns investing philosophies into something you can watch argue. Seeing a value investor and a growth investor disagree about the same earnings report is genuinely useful.

Freqtrade runs a crypto strategy from start to finish: entry rules, stop losses, trailing stops, profit targets and leverage. It tests on history with fees included, tunes your settings, and runs in dry-run mode with fake money before you go live.

Its FreqAI add-on retrains machine learning models on recent data so a strategy can adapt as markets change. You can check on it from a web dashboard or a Telegram bot on your phone.

It is the rare free tool that is both serious and maintained almost daily. That is why it tops our Community Score at 96.

Qlib covers the whole quant workflow: storing price data, building factors (signals that may predict returns), training machine learning models, and testing portfolios. It ships with large ready-made factor sets so you are not starting from zero.

It is built for research. The output is signals and simulated portfolios, not live orders.

Microsoft's name, a complete pipeline and academic credibility. Many people learning machine learning for finance start here.

Every crypto exchange has its own API with its own quirks. CCXT gives your code one consistent way to fetch prices, check balances and place orders across all of them.

You rarely see it, but it sits underneath many of the crypto bots on this list, including Freqtrade.

It removes the most boring, error-prone work in crypto automation. Plumbing that just works earns stars quietly for years.

Vibe-Trading gives a team of AI agents a full research toolkit: market data tools, finance skills and several backtest engines. Agents can research an idea, test it on history and analyze a trade journal.

It is one of the most active projects in the AI agent lane, with updates landing constantly.

It goes further than most agent demos by pairing the AI team with real backtesting, so ideas get tested instead of just discussed.

Nautilus Trader runs your strategy on history and in live markets with the exact same code. That removes a nasty category of surprises where a strategy behaves one way in testing and another way with real money.

Its core is written in Rust for speed, and you drive it with Python.

Serious quants love it because it was built for production from day one. Our community research came back positive, and it scores 92.

Backtrader replays your strategy on historical prices with realistic fees, slippage and order types. It is readable, well documented and covered by countless tutorials.

It searches about 1,600 times a month in the US, more than almost any repo here, which tells you how many people still learn on it.

It made backtesting approachable for Python beginners years before most alternatives existed.

AI-Trader lets agents like Claude Code, Codex and Cursor join a trading platform through a single skill file. The agents read news, central bank speeches and earnings tone, then act on what they conclude.

It also lets agents share ideas, and lets people follow what the agents do.

It points at where things are heading: AI agents reading the qualitative stuff numbers miss, like tone and context.