TradingAgents: how do 12 AI agents discuss a single stock?

Imagine you have to decide about one stock. At a good trading firm one person doesn't settle that alone: someone reads the reports, someone looks at the chart, someone objects and says "no, this is risky".

TradingAgents repeats exactly that process with AI. The README lists the teams, and we opened the code and counted the agents one by one. There are 12. We didn't run it, so here we only talk about what each one is meant to do.

Four analysts

They work at the same time and each writes a report:

• The fundamentals analyst reads financial documents, looking at whether the company is healthy and for red flags.
• The market analyst looks at the chart and picks up to 8 indicators for the situation, such as RSI and MACD.
• The news analyst covers company and world news from the past week.
• The sentiment analyst gathers headlines, StockTwits and Reddit chatter. Its instructions contain a nice line: don't confuse an event with someone's opinion.

The debate: the best part

Then two researchers take opposite sides. The bull argues why the stock could rise, the bear why you shouldn't buy it. They don't just read out lists, they answer each other's points.

It's a good idea, because people tend to decide first and then look only for information that agrees. Here the opposing view is built into the system on purpose. The research manager closes the debate and hands the trader a clear plan.

The trader and the risk team

The trader reads the plan and proposes buying, selling or waiting, including when and how much.

Then three risk analysts argue: one chases the bigger gain, one protects the money, and the third looks for balance between them. Three views of the same data.

The last word

The portfolio manager decides. The result is one of five levels: buy, overweight, hold, underweight or sell. According to the README an approved order goes to a simulated exchange, and we found no module in the repo that connects to a real broker.

Is the project alive?

Releases answer that better than stars. There have been 15 since February, with three each in September and October:

Its popularity isn't in doubt either. Among the projects we looked at for AI Tools, this one has the most stars:

What we think

This is a tool for developers. You need Python 3.11 and an AI model key, and every analysis costs model usage. For an ordinary investor the most useful part isn't the code, it's the process: hear the opposing view before deciding, and look at risk from more than one side.

You can do that without AI. Next time you want to buy a stock, try writing down three reasons why you shouldn't.

The authors say it plainly too: this is a research tool, not financial advice. Results depend on the model, the period and the quality of the data.

For information only, not investment advice. This is a third-party project and we're not affiliated with it.