TradingAgents
A research framework staging several agents with distinct roles, analysts, researchers and risk managers, and having them debate before producing a buy or sell decision. Derived from a research paper, under Apache 2.0. You need a paid model key and a market data key. It is not an order execution engine, and results are not reproducible from one run to the next.
Strengths
- A readable implementation of debate between opposing roles before a decision, transferable outside finance
- Apache 2.0 licence and backing from a research paper, so a documented method
Limitations
- Non-reproducible outputs: two runs on the same asset on the same day diverge
- The model can invent prices, something recent fixes are still working to contain
- Non-trivial cost per analysis: each decision makes several agents debate, multiplying calls
Best for
- Studying adversarial debate between agents to transfer it to a committing product decision