Contribution Proposal: Azure OpenAI Provider Support

[zantis]

Azure OpenAI Provider Support for Zed

Problem Statement

Currently, Zed lacks support for Azure OpenAI. Many enterprises mandate Azure OpenAI over OpenAI due to compliance, data governance, and security requirements. Given that my organization relies on Azure OpenAI, I developed this support out of necessity and would now like to contribute it back.

I recognize there may be some discussion overlap and duplication, but followed the contribution guidelines encouraging discussion here prior to making a pull request

Community Demand

This has been a frequently requested feature with community interest and discussion:

• Open Feature Request: Azure OpenAI support #12296
• Recurring Community Discussions:
  • Is it possible to use Azure OpenAI API key for Assistant? #10131
  • How does one configure Azure Open AI for assistant? #16444
  • How to configure Azure OpenAI #20232
  • Azure AI Foundry support #25544

Solution Proposal

I have implemented and thoroughly tested a dedicated Azure OpenAI provider. Here's how it works:

Open the Agent Panel → Configure → Enter your API key
Configure your model deployments in your Zed settings ~/.config/zed/settings.json. Example is shown in the panel for convenience
Select your Azure OpenAI model in the Agent Panel
gpt-4.1 (write) example
o4-mini (ask) example
o1 (ask) example

File Changes & Components

• Provider Registration: Added and registered the new Azure OpenAI provider → crates/language_models/src/language_models.rs
• Module Declaration: Added azure_openai module → crates/language_models/src/provider.rs
• Core Implementation: Nearly all the effort focused in this new file azure_open_ai.rs. By reusing the existing OpenAI crate, minimal additional development was required → crates/language_models/src/provider/azure_open_ai.rs [NEW]
• Settings Support: Added Azure OpenAI provider settings support → crates/language_models/src/settings.rs
• Agent Integration: Added azure_openai as a valid provider option for default model selection → crates/agent_settings/src/agent_settings.rs
• UI Icon: Registered new AiAzureOpenAi icon → crates/icons/src/icons.rs
• Azure OpenAI Icon: assets/icons/ai_azure_open_ai.svg [NEW] as shown in the screenshots above. (icon used to the right of this arrow) →
• Documentation: Added Azure-specific configuration documentation → crates/language_models/src/configuration.md

(begin documentation snippet)

Azure OpenAI {#azure-openai}

✅ Supports tool use

Zed supports Azure OpenAI models through the Azure AI Foundry or Azure Portal.

Setting up Azure OpenAI

1. Create an Azure OpenAI resource in Azure AI Foundry or via the Azure Portal
2. Deploy one or more models in your Azure OpenAI resource
3. Obtain your API key and endpoint from the resource's "Keys and Endpoint" section
4. Open the settings view (agent: open configuration) and navigate to the Azure OpenAI section
5. Configure your Azure OpenAI API key then complete your Zed settings.json to set up the Azure OpenAI models you want to use

Azure OpenAI Configuration

To configure Azure OpenAI models, add the following to your Zed settings.json:

  "language_models": {
    "azure_openai": {
      "resource_name": "your-resource-name",
      "api_version": "2025-04-01-preview",
      "available_models": [
        {
          "name": "gpt-4.1",
          "deployment_name": "your-gpt-4.1-deployment-name",
          "display_name": "GPT-4.1 (Azure)",
          "max_tokens": 32768
        },
        {
          "name": "o4-mini",
          "deployment_name": "your-o4-mini-deployment-name",
          "display_name": "o4-mini (Azure)",
          "max_tokens": 32768,
          "max_output_tokens": 32768
        }
      ]
    }
  }

Configuration Fields

• resource_name: The name of your Azure OpenAI resource (not the full endpoint URL)
• api_version: Azure OpenAI API version
• deployment_name: The deployment name you created in Azure AI Foundry or Azure portal
• name: The underlying OpenAI model name (must match the model used in your deployment)
• max_tokens: Maximum context window size for the model
• max_output_tokens: Maximum number of tokens the model can generate in a single response
• max_completion_tokens: Alternative to max_output_tokens used specifically by o-series reasoning models

You must provide the model's Context Window in the max_tokens parameter; this can be found in the Azure OpenAI model documentation.
Azure OpenAI o-series models should set max_completion_tokens as well to avoid incurring high reasoning token costs.

(end of documentation snippet)

Considerations

This enhancement seems to align with the broader goals of the Zed to smoothly integrate LLMs in multiple ways across the editor. I'm excited to contribute this to Zed and help make it more accessible to enterprise developers. This is my first time contributing to Zed, so I welcome any feedback on both the technical approach and contribution process. Appreciate your consideration and look forward to your feedback. Thanks.

-Z

[linozen]

I'd very much like to see this in Zed.

[sanzoghenzo]

I also would like to see this!

One thing, though: does this support custom domains/base URLs?

Also, it would be great to be able to set the provider and models configuration through UI like the OpenAI compatible provider:

I recognize there may be some discussion overlap and duplication, but followed the contribution guidelines encouraging discussion here prior to making a pull request

It seems to me that there's enough discussion to go ahead and make the PR. 😉
CONTRIBUTING.md states:

The best way to get us to take a look at a proposed change is to send a pull request.

[Jaakkonen]

Support for Microsoft Foundry Azure OpenAI models was implemented in #43742 but Zed team closed the PR saying they will enable specifying models via extensions soon™. There's a AWS Bedrock provider related PR in #41393 so let's see what happens with it.