Samuel Colvin

I guess we could add a `pydantic-core` type `Unset` which `exclude_unset` respected.

Thanks the policy is here https://ai.pydantic.dev/contributing/#new-model-rules

PR welcome to add AWS Bedrock. I think from a brief search that Bedrock is like VertexAI - you use a custom API, but the format of the response matches...

We will support Claude via bedrock, as it should just be setting the url and configuring auth. Not sure about other models. PR welcome to try to implement it, otherwise...

Is this really necessary?

Well, I guess if the model is able to return structured responses of some kind, you could implement a custom `Model` which somehow parses the response and builds instances of...

I think I agree it's just a different way of building with LLMs, happy to consider a specific case where we use or that library or learn from it.

Closing, but happen to reopen if there's something concrete we can do.

I think it's worth having a dedicated model, even if it's just setting the base url. Also, look like we'll need more specific support for Ollama, see #242.

Makes sense, the biggest challenge is rendering the API docs in markdown, see https://github.com/mkdocstrings/mkdocstrings/issues/701.