crewAI stock analysis with Ollama

My crew had a Financial Analyst, a Staff Research Analyst and an Investment Advisor, and the input was "Analyze BOX".

Llama2 often lost context and failed to use the right tools.

Mistral used all the tools and collaborated with the other agents, though it sometimes scraped the wrong articles.

I hosted the Ollama server on a RunPod GPU Pod with an A40 at $0.69 an hour.

I also tried Gemma and wouldn't use it again, especially with quantized models.

My ranking is Mistral > Llama2 > Gemma.