support local llm server

[jianlins]

No description provided.

[pachieh]

Agree, adding support for Ollama would be amazing. Is it just about changing the base_url to whatever my Ollama server URL is?

[Jyun1998]

Agree, adding support for Ollama would be amazing. Is it just about changing the base_url to whatever my Ollama server URL is?

no it's not. API format is quite different as well as model option params.

I might make pr for this

[pachieh]

Please do. AgentLab would be super useful air gapped networks where we can't/won't run OpenAI or DeepSeek but have Ollama instances running.

[jianlins]

Agree, adding support for Ollama would be amazing. Is it just about changing the base_url to whatever my Ollama server URL is?

Yes. it can be. You can start with "ollama serve." And then use openAI API to communicate with it. But I found using llama.cpp directly would be more helpful, you can configure the parameters when you start the "llama-server"

[yanyuechuixue]

I think line 193, 196 and 199 in agents.py should be self.base_url instead of self.base.