TradingAgents GUI setup with Ollama, GPT or Claude

The Local Lab presents TradingAgents GUI, the speaker's browser interface for a trading research framework originally operated through a Python CLI. Its simulated desk combines analyst reports, bull and bear debates, a trade plan and risk reviews before producing a buy, sell or hold recommendation. The speaker cautions that LLMs can hallucinate and describes the output as a second opinion rather than financial advice.

The setup walkthrough covers Windows, Mac and Linux with Python 3.10 or newer and Git. After cloning the fork, users run the platform's install script, then its start script to open a local web server at localhost:5000. Provider options include Claude Opus 4.7, GPT 5.5, OpenRouter and Ollama for those who want to run models locally. The speaker warns that smaller models can struggle with complex agent steps and suggests models around 70 billion parameters when VRAM permits. Local inference still accompanies live market data, which the GUI retrieves from Yahoo Finance by default, with Alpha Vantage as a backup.

The demonstration covers separate quick and deep thinking models, analyst selection, debate depth and report length. Users can track token usage and estimated costs, export reports as Markdown, HTML or PDF, and pin saved analyses as context for follow-up chat. A Patreon Windows installer is also mentioned.