Are there any relatively complex open-

Open source LLMs low-key suck at being agents (and really no LLM is good at agentic behaviour yet, minus like gpt-4-turbo minimum)

But, you can use any LLM with a react agent, and it will do ok-ish. I would use ollama

This notebook uses ollama, but ends up using llama3-70b over replicate for the agent section because the extra horsepower is kind of needed to make a reliable agent
https://docs.llamaindex.ai/en/latest/examples/cookbooks/llama3_cookbook_ollama_replicate/?h=llama3#7-agents

I mean like an open source application, not model