Incompatible API calls when using Ollama locally

[dvirla]

Initial Checks

• I confirm that I'm using the latest version of Pydantic AI
• I confirm that I searched for my issue in https://github.com/pydantic/pydantic-ai/issues before opening this issue

Description

I am trying to run Qwen3 via Ollama locally and getting api call errors (400) as presented below:

Exception has occurred: ModelHTTPError
status_code: 400, model_name: qwen3:32b, body: {'message': 'invalid message content type: <nil>', 'type': 'invalid_request_error', 'param': None, 'code': None}
httpx.HTTPStatusError: Client error '400 Bad Request' for url 'http://localhost:11434/v1/chat/completions'
For more information check: https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/400

During handling of the above exception, another exception occurred:

openai.BadRequestError: Error code: 400 - {'error': {'message': 'invalid message content type: <nil>', 'type': 'invalid_request_error', 'param': None, 'code': None}}

The above exception was the direct cause of the following exception:

  File "/data/home/dvirla/llm_search_eval/src/services/base_agent.py", line 56, in run
    response = self.agent.run_sync(user_input)
  File "/data/home/dvirla/llm_search_eval/src/services/base_agent.py", line 65, in test_agent
    result = agent.run(user_query)
  File "/data/home/dvirla/llm_search_eval/main.py", line 5, in <module>
    res = test_agent()
pydantic_ai.exceptions.ModelHTTPError: status_code: 400, model_name: qwen3:32b, body: {'message': 'invalid message content type: <nil>', 'type': 'invalid_request_error', 'param': None, 'code': None}

The code to reproduce is attached.
I have tried multiple ways to initialize the agent as detailed in the documentation, but still getting this error.
When I try to run the model from the terminal using Ollama, I see it uses the endpoint /api/generate and not v1/chat/completion.

Example Code

class BaseAgent:
    def __init__(self, prompt_path: str = None, provider_name: str = "Google", model_name: str = "gemini-flash-latest"):
        if provider_name == "Google":
            settings = GoogleModelSettings(google_thinking_config={'include_thoughts': True})
            provider = GoogleProvider(api_key=os.getenv("GOOGLE_API_KEY"))
            self.model = GoogleModel(model_name, provider=provider)
        elif provider_name == "OpenAI":
            provider = OpenAIProvider(api_key=os.getenv("OPENAI_API_KEY"))
            self.model = OpenAIChatModel(model_name=model_name, provider=provider)
        elif provider_name == "ollama":
            provider = OllamaProvider(base_url='http://localhost:11434/v1/')
            self.model = OpenAIChatModel(
                model_name=model_name,
                provider=provider
            )
        else:
            raise ValueError(f"Invalid provider: {provider_name}")

        if prompt_path:
            self.system_prompt = self._load_prompt(prompt_path)
        else:
            self.system_prompt = [None]

        self.agent = Agent(
            model=self.model,
            system_prompt=self.system_prompt,
            output_type=str
        )

    @staticmethod
    def _load_prompt(path: str) -> str:
        with open(path, 'r') as f:
            return f.read()
    
    def run(self, user_input: str):
        response = self.agent.run_sync(user_input)
        return response

def test_agent():
    agent = BaseAgent(
        provider_name="ollama",
        model_name="qwen3:32b"
    )
    user_query = "What is the capital of France?"
    result = agent.run(user_query)
    print(user_query)
    print(result)

Python, Pydantic AI & LLM client version

ollama version is 0.12.6
pydantic_ai 1.14.1
Python 3.11.4

[DouweM]

@dvirla I expect that the issue is in self.system_prompt = [None], as the system_prompt argument to Agent is required to be a string or sequence of strings -- no None allowed:

        if prompt_path:
            self.system_prompt = self._load_prompt(prompt_path)
        else:
            self.system_prompt = [None]

        self.agent = Agent(
            model=self.model,
            system_prompt=self.system_prompt,
            output_type=str
        )

I suggest using a type checker like pyright, manually or integrated in your IDE, to help catch issues like this.

Please let me know if the issue still occurs without that None.

[dvirla]

Worked, thanks!