TradingAgents setup with Ollama, Gemma 4 and Qwen

This tutorial builds a local AI stock research workflow with TradingAgents, a framework the presenter describes as open source and built on LangGraph. Analyst teams examine fundamentals, sentiment, news and technical indicators. Bullish and bearish researchers debate their reports before trader and portfolio management agents assess a proposed decision.

Ollama runs Gemma 4 E2B and Qwen 3.5 4B. The presenter says the smaller Gemma model can run with 8GB RAM and requires about a 7GB download. A custom Ollama model receives instructions to omit thinking output for the demo; the tutorial does not establish that these instructions disable internal reasoning.

The Python setup configures model roles, debate rounds, stock tickers and analysis dates. It accepts a ticker and date through command-line arguments, or processes the configured batch, then saves decisions and reasons. The Reliance example ends with a hold proposal. The presenter treats it as a demonstration rather than an investment recommendation.

Local inference is distinct from offline market research here: the workflow uses yfinance to fetch stock prices and discusses external news. The presenter also describes persistent local memory and checkpoint recovery. Broker execution, extra technical indicators and a dashboard are suggested extensions.