Ollama support

[cal859]

fix #112.

A PR to implement an OllamaModel and allow for easy access to Ollama.

A few questions / notes:

• The main issue with Ollama is the typing of the model names. Ollama has 100's of different models and many more are added daily/weekly, so maintaining a complete list (like for Gemini) seems like it will be quite difficult. It is possible to pull the models available to the current user, using ollama.list(), and I have put a (commented out) example of how this would work in the models/ollama.py file, but for now, I have just left as str as the type
• The second issue with the above typing is, because there is so much overlap with this code and the openai code, I didn't want to completely replicate the code between them, so I am using the existing OpenAIAgentModel, rather than reimplementing that, but then to get that work with the Ollama model, I have had to allow a much more general str type as well as the much more tightly defined ChatModel type. If you think it makes sense to just fully duplicate the OpenAIAgentModel for an OllamaAgentModel and then can have more strict typing for both versions, then just let me know

Otherwise, this works well.

To test it:

• Follow the setup instructions added to docs/install.md#ollama, install the qwen2.5-coder:latest model, and then run:

PYDANTIC_AI_MODEL=ollama:qwen2.5-coder:latest uv run -m pydantic_ai_examples.pydantic_model 

[hyperlint-ai]

The style guide flagged several spelling errors that seemed like false positives. We skipped posting inline suggestions for the following words:

• Ollama

[samuelcolvin]

Amazing! I'm just getting on a flight back to the UK, if the Internet is usable, I'll review soon.

[samuelcolvin]

Otherwise this looks great! 🎉

Really looking forward to releasing this.

[cal859]

Thanks for your review @samuelcolvin .

I have addressed the majority of issues, I just had a couple of questions (left on the unresolved comments)

EDIT: I'm investigating the failing CI/lint check. It is working locally

[samuelcolvin]

you need to merge with/rebase against main to get tests passing, do you want me to help?

[hyperlint-ai]

The style guide flagged several spelling errors that seemed like false positives. We skipped posting inline suggestions for the following words:

• Pydantic
• Logfire
• FastUI
• [Cc]ollab

[cal859]

you need to merge with/rebase against main to get tests passing, do you want me to help?

@samuelcolvin thanks for the advice. After getting into a bit of a pickle I think I have successfully rebased now. Everything seems to be running successfully. I've made all of the additional changes you have requested, and the following command still works when testing the ollama model. Let me know what else is required if anything before merging.

PYDANTIC_AI_MODEL=ollama:llama3.2 uv run -m pydantic_ai_examples.pydantic_model

[samuelcolvin]

This is awesome, I'll create a new PR to add tests since they require special permissions to run.

Thank you so much.