@crmne it doesn't quite work for me.

To make RubyLLM work with Azure OpenAI, I added support for a custom chat/completions path, since Azure uses a different URL structure.

What was done:

1. Added a `completion_url` method to `RubyLLM::Providers::OpenAI`:

def completion_url
  RubyLLM.config.completion_url || 'chat/completions'
end

2. Added the `completion_url` option to the RubyLLM configuration:

RubyLLM.configure do |config|
  config.openai_api_key = "your-azure-api-key"
  config.openai_api_base = "https://test-east-2.openai.azure.com"
  config.completion_url = "openai/deployments/gpt-4o-mini/chat/completions?api-version=2025-01-01-preview"
end

Key differences:

Unlike the standard OpenAI API, Azure determines the model not through the model parameter, but via the request URL (deployments/<name>) and a required api-version query parameter.
This means that dynamically switching models using the model: option isn’t feasible, since different deployments may require different API versions.

Conclusion:

To support Azure OpenAI in RubyLLM, it's necessary to bypass the internal model abstraction and manually define the request path. While this approach works, it limits flexibility when switching between models.

I didn't do a PR yet because it was a minute decision to just check that the main problem is this and not something else.

I guess it should be done by a separate provider, maybe I can do it, I look forward to your opinion, thanks.

What was done:

1. Added a completion_url method to RubyLLM::Providers::OpenAI:

2. Added the completion_url option to the RubyLLM configuration:

Key differences:

Conclusion:

1. Added a `completion_url` method to `RubyLLM::Providers::OpenAI`:

2. Added the `completion_url` option to the RubyLLM configuration:

Uh oh!

Azure OpenAI support #15

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions