This pull request introduces support for a new provider type, openai-responses, which enables integration with OpenAI-compatible Responses APIs. This has been tested using GPT Codex models in Azure AI Foundry, as these do not support using the Chat Completions API. The existing OpenAI compatible provider was used as a base, so functionally it should be identical except for the API being used.

All code was written using Kilo Code with GPT-5.2-Codex.

Core provider and configuration support

• Added openai-responses to provider lists, schemas, and validation logic in provider-settings.ts, ensuring it is recognized as a custom provider and its settings are properly validated and merged. [1] [2] [3] [4] [5]
• Updated model, label, default model, and required fields mappings to include openai-responses in relevant constants. [1] [2] [3] [4] [5]
• Ensured default model selection and model info logic supports openai-responses alongside other OpenAI providers. [1] [2] [3]

API handler and backend integration

• Implemented OpenAiCompatibleResponsesHandler for the new provider, integrated it in the API handler factory, and added unit tests for its behavior and streaming fallback. [1] [2] [3] [4]

UI and settings integration

• Added openai-responses as a selectable provider in the settings UI, ensured proper handling of its configuration fields, and updated logic for custom headers and slug mapping. [1] [2] [3] [4]
• Updated utility functions to correctly fetch API key and base URL for openai-responses profiles. [1] [2]

Internationalization

• Added new translation keys for "OpenAI Compatible (Responses)" provider in multiple languages to support UI updates. [1] [2] [3] [4]