Setup/HowTo for using Agent Mode?

[webbben]

Hi, I'm running Odysseus on docker, on macOS. I'm having some issues getting agent mode to work. It seems likely that I'm doing something wrong, but I'm not sure where to look for instructions/requirements for using this feature. Perhaps my model doesn't support it, or something like that, but unsure.

Anyway, I'm starting with just debugging the issues I have.

(BTW - I take it that discussions are a better place to debug things than the issues area? Seems like issues is meant less for getting help debugging things, and more for reporting something you strongly suspect of being a bug).

I'm running Ollama, and when I try to get Qwen:3.4b to use agent mode, this is what happens:

The first basic issue that jumps out is:

GET /api/files/list -> HTTP 404

So that looks to me like one of the API endpoints is not reachable for some reason. I'm not really sure where to start with actually debugging this further, but I found what I think is the corresponding logs in Docker:

INFO:     127.0.0.1:48416 - "GET /api/files/list?path=%2FUsers%2Fbenwebb%2Fdev%2Fai-playground HTTP/1.1" 404 Not Found

2026-06-05 00:44:31,565 - httpx - INFO - HTTP Request: GET http://localhost:7000/api/files/list?path=%2FUsers%2Fbenwebb%2Fdev%2Fai-playground "HTTP/1.1 404 Not Found"

2026-06-05 00:44:31,566 - src.tool_execution - INFO - Tool executed: app_api -> exit_code=1

Do I just not have something configured or set up correctly to use this API? Anything I should check first?

Also, if this is just a matter of my LLM model not supporting this stuff, is there a good, standard way to confirm which models work for this stuff? Would be cool if Odysseus could even show a warning to you that your LLM model may not be compatible.

[jaimebarrancos]

Hey,

• Yes certain models don't work well / weren't trained to use functions (mostly smaller models).
• Qwen does supports calling functions well.

I have similar issues with using skills and memories for example. I'm using Qwen3.5 2B

[ejangggg]

What model do you recommend? ive tried gemme4, qwen and gpt oss and they just act like chatbots

[jaimebarrancos]

Which specific models are you using and what are their parameter sizes?

[jorgeporragas]

I've run into this issue a lot. Agent mode simply doesn't work well, unfortunately. Not yet anyways. It often goes into loops replying to itself and trying to run commands I didn't ask it to. I asked it to look up steps how to set up a GitHub repository (to test the agent and web search) and it started to try to create a GitHub repo in my machine. I don't think it's an agent issue because this has happened with different models. I still need to run some more tests.

[jaimebarrancos]

I believe discussion #2724 is also working on this same problem.