Support Azure OpenAI integration

[dawnwages]

Could we support the integration of Azure OpenAI in this repo? Sounds like an awesome alignment!

https://github.com/Azure/azure-openai-samples

[dawnwages]

@pamelafox

[pamelafox]

I work on many of our Python Azure OpenAI samples (like https://github.com/Azure-Samples/azure-search-openai-demo/) and would be happy to help with this. Let me know if this is a feature you're looking for in this repo.

[tomusher]

Thanks both - this sounds great and definitely something we'd be interested in integrating.

There's two main areas I think Azure OpenAI could be integrated within the current state of our AI packages:

• Using Azure's OpenAI chat endpoints for the wagtail-ai rich text integration. This is currently built using llm, allowing us to use a plugin-based interface for adding support for different AI models. A plugin for Azure's OpenAI API for llm would make it easy for wagtail-ai users to use them.
• Our other AI package; wagtail-vector-index is focused on providing developer tools and AI search using AI APIs. Adding Azure AI Search as a vector backend to this would be great.

There's loads of other nice things that Azure OpenAI can do that we'd like to consider when the packages matures/integrates more AI-oriented features.

[pamelafox]

Update: We tried wagtail-ai in a live stream. I was hoping we'd be able to use it with Azure OpenAI, as llm does actually support Azure OpenAI, but wagtail-ai would need additional options to get it to work. I'm curious why you went with llm? I love it as a CLI tool, but it doesn't seem optimal for a deployed server tool, given all its reliance on file system paths. I would think something like langchain would be more configurable in a deployed app.

[pamelafox]

Re wagtail-vector-index: very cool! I've shared that with another advocate who's been working on improving the AI search support in other OSS packages, to see if they'd have time to add Azure AI search.

[tomusher]

Update: We tried wagtail-ai in a live stream. I was hoping we'd be able to use it with Azure OpenAI, as llm does actually support Azure OpenAI, but wagtail-ai would need additional options to get it to work. I'm curious why you went with llm? I love it as a CLI tool, but it doesn't seem optimal for a deployed server tool, given all its reliance on file system paths. I would think something like langchain would be more configurable in a deployed app.

Thanks for giving it a go @pamelafox! I'd love to hear more about what additional options we're missing to get Azure OpenAI working/what roadblocks you ran in to with llm.

On the llm choice - we wanted something simple that was easily extendable by developers (through llm plugins) and that didn't come with all the extra baggage that Langchain provides. Simon was interested in feedback on it over on issue #56 if you wanted to add any additional comments.

Saying that; we're not tied to llm. The swappable backend architecture is designed so we can switch out with a potential future LangchainBackend, or perhaps even a more specific AzureOpenAIBackend. You might have noticed we've recently added an OpenAIBackend as we experiment with multi-modal models and start running in to some of the limitations of the llm Python API.

Really appreciate your feedback - we'd still love to make sure we fully support Azure OpenAI.