Fix Azure/OpenAI endpoints and reject Azure AI Inference URLs

[Neonsy]

Context

The OpenAI Responses provider had endpoint-shape mismatches when users supplied Azure-style URLs, especially deployment/chat-completions paths. This caused fallback behavior to hit incorrect URLs and mixed auth/query behavior between OpenAI and Azure. We also needed to fail fast for unsupported Azure AI Inference endpoints (*.services.ai.azure.com) and tighten config validation for openai-responses.

Implementation

• Added explicit endpoint classification for Azure OpenAI vs Azure AI Inference in the Responses provider.
• Added a guard that rejects Azure AI Inference endpoints with a clear actionable error message.
• Normalized fallback base URL construction so requests resolve cleanly to /responses without duplicating or mis-shaping /v1.
• Updated fallback auth behavior:
  • Azure OpenAI fallback uses api-key header.
  • Non-Azure fallback uses Authorization: Bearer ....
• Added Azure-specific api-version behavior only when needed for non-/openai/v1 paths.
• Improved 404 guidance for Azure users, including warning against /deployments/.../chat/completions as a base URL.
• Added openai-responses validation in webview settings (openAiBaseUrl, openAiApiKey, openAiModelId required).
• Added targeted tests for URL normalization, Azure auth/query behavior, unsupported endpoint rejection, and validation behavior.

How to Test

• Run provider tests:
  • cd src && pnpm test api/providers/__tests__/openai-responses.spec.ts
• Run webview validation tests:
  • cd webview-ui && pnpm test src/utils/__tests__/validate.spec.ts
• Manual sanity checks:
  • Configure openai-responses with Azure OpenAI base URL https://<resource>.openai.azure.com/openai/v1 and deployment model name; confirm requests target /openai/v1/responses.
  • Configure an Azure AI Inference URL (https://<resource>.services.ai.azure.com/...); confirm the provider fails immediately with the unsupported-endpoint error.
  • Use a deployment/chat-completions-style Azure URL and confirm fallback normalization still targets /openai/v1/responses correctly.

Note

The Discord user who has brought it up tested this change.

[changeset-bot]

🦋 Changeset detected

Latest commit: 0e10528

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
kilo-code	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[Anthonix24812]

Can confirm that this fix works with a Microsoft Foundry deployment of gpt-5.2-codex with Model version 2026-01-14

[Neonsy]

@kevinvandijk

[kevinvandijk]

Thanks! Looks good to me

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Overview

This PR refactors Azure endpoint handling in the OpenAI Responses provider to:

1. Reject Azure AI Inference endpoints (*.services.ai.azure.com) early with a clear error message, instead of attempting to use them with unsupported api-version query parameters.
2. Normalize Azure OpenAI base URLs for the fallback fetch path, stripping deployment-specific paths and api-version params to produce clean /openai/v1/responses URLs.
3. Use api-key header (instead of Authorization: Bearer) for Azure endpoints in the fallback path.
4. Add openai-responses validation in the webview settings to require base URL, API key, and model ID.

The URL normalization logic in normalizeResponsesBaseUrl correctly handles all tested Azure URL patterns (deployment URLs, cognitiveservices, bare endpoints, /v1 paths). The shouldAppendAzureApiVersion method ensures /openai/v1 paths don't get unnecessary api-version query params.

Test coverage is thorough — new tests cover Azure AI Inference rejection, Azure OpenAI SDK path options, Azure fallback auth headers, deployment URL normalization, cognitiveservices endpoint normalization, and validation for the openai-responses provider.

Files Reviewed (5 files)

• .changeset/azure-responses-v1-endpoints.md - Changeset
• pnpm-lock.yaml - Lock file (skipped)
• src/api/providers/openai-responses.ts - Core implementation changes
• src/api/providers/__tests__/openai-responses.spec.ts - Test updates
• webview-ui/src/utils/validate.ts - Validation for openai-responses provider
• webview-ui/src/utils/__tests__/validate.spec.ts - Validation tests