TradingAgents setup with OpenRouter and Ollama support

TradingAgents is an open source financial analysis framework built around specialized LLM roles. The speaker explains four layers: analysts gather market and company data, bull and bear researchers debate it, a trader proposes a position, and risk agents review that proposal before a portfolio manager receives it. Claims that this structure eliminates hallucination or agreement bias are the speaker's assertions, not demonstrated guarantees.

The tutorial covers version 0.2.4 and describes LangGraph checkpointing, structured JSON outputs, and support for DeepSeek, Qwen and Ollama. A local LLM option is discussed, but the demonstrated setup uses OpenRouter with a Finnhub API key. Running the framework on your own machine therefore does not make this example offline.

The walkthrough clones the repository, prepares a Python virtual environment, installs dependencies and configures a .env file. It launches the interface with python -m cli.main, selects AAPL, enables four analysts and chooses shallow research. The transcript abbreviates some installation commands, so it does not provide a complete command reference.

The speaker reports a 26.6% cumulative return in a three-month Apple benchmark, then cautions that a short backtest can overfit. Other limits include roughly 11 LLM calls and more than 20 tool calls per decision, API costs, and duplicate or delayed news data that needs cleaning.