[Feature]: Support / documentation for Azure OpenAI configuration

[Wendyfff0616]

Which destkop app does this feature request relate to?

• [ ] Select a project 👇
• [x] Agent-TARS (cli, server, agent, tool etc.)
• [ ] UI-TARS Desktop

What problem does this feature solve?

I noticed that the CLI workspace wizard (agent-tars workspace --init) lets me pick “azure-openai” as the default model provider, but I couldn’t find any docs or examples that show:

1. Whether Azure OpenAI is officially supported right now.
2. What a minimal agent-tars.config.ts (or JSON/YAML) should look like for Azure OpenAI—including required fields like endpoint, apiKey, deployment name, etc.
3. Any CLI-only flags I may need (e.g. --azure.endpoint vs --model.baseURL) .

What does the proposed features look like?

Steps I tried

agent-tars workspace --init   # chose “azure-openai”
# then edited agent-tars.config.ts roughly like:

export default defineConfig({
  model: {
    provider: 'azure-openai',
    id: 'gpt-4o',
    apiKey: process.env.AZURE_OPENAI_API_KEY,
  },
});

Starting with agent-tars --open results in:

InputError: Azure OpenAI endpoint is required.

If I add baseURL,

agent-tars workspace --init   # chose “azure-openai”
# then edited agent-tars.config.ts roughly like:

export default defineConfig({
  model: {
    provider: 'azure-openai',
    id: 'gpt-4o',
    apiKey: process.env.AZURE_OPENAI_API_KEY,
    baseURL: 'https://<resource>.openai.azure.com',
  },
});

Starting with agent-tars --open results in:

Error: baseURL and endpoint are mutually exclusive

Thank you!

[ulivz]

@Wendyfff0616 Thanks for your feedback! I often use azure-openai provider, but I noticed that our API serving provider are different, could you tell post the original model manufacturer’s documentation where you apply the azure-openai model, I can follow your step to reproduce it.

[Wendyfff0616]

Thanks for the reply!

I followed the standard Microsoft Azure OpenAI setup as shown in their quickstart guide. Here's a simplified reference based on the documentation I used: Get Started.md

---

By the way, I just realized how you're currently using azure-openai in the codebase—you're using a third-party API proxy service that provides Azure OpenAI-compatible API (with AWS_CLAUDE_API_BASE_URL and model aws_sdk_claude37_sonnet), rather than Microsoft's native Azure OpenAI service？

model: {
      provider: 'azure-openai',
      baseURL: process.env.AWS_CLAUDE_API_BASE_URL,
      id: 'aws_sdk_claude37_sonnet',
    },

---

[Update]

I previously encountered the following error when running Agent TARS with my Azure OpenAI configuration:

Error: baseURL and endpoint are mutually exclusive

After investigation, I found this was caused by having the AZURE_OPENAI_ENDPOINT environment variable set in my terminal session. Once I unset this environment variable, the error disappeared.

New Problem

However, I now encounter a new error:

[Stream] Error in agent loop execution: Error: 400 Unrecognized request argument supplied: thinking

Analysis

After reviewing the Agent TARS code, I guess the root cause is as follows:

In UI-TARS-desktop/multimodal/agent/src/agent/llm-client.ts, the following logic is used when creating the LLM client:

// UI-TARS-desktop/multimodal/agent/src/agent/llm-client.ts
return createLLMClient(resolvedModel, (provider, request, baseURL) => {
  // Add reasoning options for compatible providers
  if (provider !== 'openai') {
    request.thinking = reasoningOptions;
  }
  // ...
  return requestInterceptor ? requestInterceptor(provider, request, baseURL) : request;
});

Since my provider is named 'azure-openai', the thinking field is automatically added to the request payload. However, the official Azure OpenAI API does not recognize the thinking parameter, resulting in a 400 error.

If it's allowed, maybe change if (provider !== 'openai') to if (provider !== 'openai' && provider !== 'azure-openai')?

For your information, my config is now set as follows:

// agent-tars.config.ts
export default defineConfig({
  model: {
    provider: 'azure-openai',
    id: 'gpt-4o',
    apiKey: process.env.AZURE_OPENAI_API_KEY,
    baseURL: 'https://<resource>.openai.azure.com',
  },
});

Thank you for your help!

[mohammadkashif-shaikh-atos]

Hi any update on this issue ? I am facing similar issues.