Suggestion to use the OpenAI library for the Ollama provider

[renzepost]

Currently the Ollama provider makes HTTP calls to the Ollama API (the /api/chat endpoint). However, currently tool calling/automatic tool execution is not working. Tool calls are discarded from the response, and the Ollama API responses don't have tool ids, so they don't fit the AISuite model.

Ollama is partially compatible with the OpenAI API. I tested this and the OpenAI API does return tool ids, making it compatible with AISuite.

So I'd suggest to use the OpenAI library for the Ollama provider so that things like tool calling work better in AISuite. If this is OK I can look into creating a PR.

[Shedulab]

Hi renzepost, I had the same issue and I fixed it for me with the help of gpt-5.
Now I can use automatic tool calls. You can have a look at issue #253.
Regards, Huseyin.

[tushardhole]

I think the issue in exiting ollama provider is parsing of tool_calls node. May be ollama response structure changed after OpenAI compatibility, for me ollama responses did have tool calls with tool id in /api/chat, just that ollama providers tool_calls parsing and actual tool_calls structure was not in sync.

More details about issue in 17515174174

As you said, even for me it worked smoothly if i just changed to OpenAI provider, but i am testing it and trying to understand more about pros & cons of switching to openai or keeping ollama provider.