TradingAgents setup with Ollama and Qwen 3

This tutorial walks through TradingAgents, an open source research framework that uses a local LLM to produce stock recommendation reports. The speaker runs Qwen 3 8B through Ollama on an Nvidia RTX 4060 with 8GB of GPU memory and reports using a 16,000-token context length.

The setup covers cloning the repository, creating a conda environment with Python 3.13, installing the project with pip, and downloading the model through Ollama. In the CLI, the speaker selects NVDA, an analysis date, all four analysts, and shallow analysis depth. Qwen 3 serves as both the quick-thinking and deep-thinking model.

The architecture keeps fundamental, sentiment, news, and technical reports separate before bull and bear researchers debate them. A trader proposes a position, a risk team reviews it, and a portfolio manager makes the final recommendation. The speaker describes LangGraph as the workflow foundation and credits version 0.2.4 with checkpoint recovery and persistent decision memory.

The demonstrated NVDA run takes about 41 minutes and produces a buy recommendation with position sizing and stop-loss levels. These are model outputs, not established investment conclusions. The speaker notes that output quality depends on the underlying model. The exchange component is simulated; actual trading requires a broker connection that the user must implement.