Can't get Agent Zero to work with Ollama / Qwen

[MortiusMaximus]

It wasn't doing this yesterday but now all I get is "Agent 0: Message misformat, no valid tool request found." no matter what I prompt it. Then it gets stuck in a loop where it tries to figure out what is wrong with JSON message formatting. Yesterday I almost had it working but it still would get stuck in loops and it would keep saying "you've already said this. you have to send something else" or something to that effect until I stopped the agent, but it wasn't giving this error. I've never had it successfully running yet. Below is a screenshot of the full error I keep getting.

I don't know why this is only happening today using the same model. I tried downloading and using qwen3:14b but it does the same thing. Back to using qwen2.5 but still getting this error message I wasn't getting yesterday.

I am not sure which models to change in the settings. I have Ollama / Qwen2.5 set for all models except for "Embedding Model" which I left on the default huggingface one (but have also tried setting that to Ollama / Qwen2.5 as well.)

Also, I have to turn vision off or else it gets stuck in a loop where no matter what I put it says: Agent 0: Generating Agent 0: Using tool 'vision_load' Paths /path/to/image.png Copy Result No images processed Copy Agent 0: Generating Agent 0: Using tool 'vision_load' Paths /path/to/your/image.png Copy Result No images processed Copy Agent 0: Generating Agent 0: Using tool 'vision_load' Paths /path/to/your/image.png Copy Result No images processed Copy" over and over again while it keeps trying to solve this issue but I didn't prompt it with anything but a simple question and no images.

[MortiusMaximus]

It seems that Agent Zero keeps sending it's documentation to ollama which has the model confused as to why I sent the documentation.

Also, I sent a request asking for it to just reply with a statement about it working and it did. Then I asked it to make a simple landing page and this happened:

and it repeated those again to which I see highlighted in yellow "You have sent the same message again. You have to do something else!" Then it proceeded to do this:

which was the previous prompt I had given it just seeing if it could even do that without errors which it did. I have no clue what is going on but it seems to just spit random stuff at the model confusing it and then loop and now apparently it goes back to old requests. I would love to get this working and it seemed to be straightforward and it seems to work great for others so any clues on why it's behaving so strangely for me? Appreciate any help on this, really would like to get this working. For how it is installed I followed instructions that included using docker desktop and I have ollama working (and it works fine prompting the model myself) and the only things I changed in agent-zero was where it had the models set to OpenAI, I changed to Ollama and typed in qwen2.5 (and qwen3:14b at one point, which seemed to make things a lot worse). If there is some other setup or something I need to do please let me know. Every time it tries to use any kind of tool it always fails with some random error and then it gets stuck in a loop.

[atrix228]

i have the same issue too after i updated to the latest version

[sgt-Script]

i have the same issue as well. :) same error as op

[frdel]

You will probably need to increase the default context window size parameter in ollama. The default used to be 2k tokens, which is not enough even for system prompt alone.