feat: OpenAI responses api #4442

kommerzienrat · 2025-12-13T23:36:17Z

Added support for OpenAI Responses API in OpenAiHandler, including handling of base URLs and API modes.
Introduced utility functions for URL normalization and Azure host detection.
Created a new module for handling Responses API requests and streaming responses.
Updated OpenAI settings UI to allow users to select between Completions and Responses API modes.
Added tests for URL utilities and Responses API functionality.
Enhanced error handling for OpenAI API interactions.

Context

It was not possible to use the latest OpenAI (compatible) models no more, e.g. GPT-5.1-codex-max, since the Responses API was not implemented in KiloCode.
Until recently, the Chat Completions API worked, but now throws a 400 error: "The chatCompletion operation does not work with the specified model". Turns out, these new Codex models (GA since Nov 13) only support the Responses API (/openai/v1/responses) now – chat fallback is gone.

Fixes #4364

Implementation

I rewrote all files related to OpenAI and OpenAI compatible calls. We now have an option to select whether the Chat Completions API or Responses API should be chosen for model calls. Since Azure and others started to save model outputs but allow us to disable this "feature", I added a checkbox for output storage. I also added tests and tried make the app more robust.

Here a generated summary:
StaticSettingsService.ts: Hardened static settings parsing—now decodes base64, JSON-parses with safeParse, funnels all failures through logAndThrow, and distinguishes Zod schema errors. This prevents silent misconfigurations and returns clearer errors.
provider-settings.ts: Added schema fields for selecting OpenAI API style (openAiApiMode) and Responses routing (openAiResponsesMode, openAiResponsesStoreEnabled). Required to validate the new UI/handler options.
openai-url-utils.ts (new): Added URL helpers to enforce https, detect Azure hosts, normalize Responses API base URLs (strip /responses, ensure /openai/v1//v1, drop api-version), and extract hosts. Supports correct client construction for Responses mode.
openai-url-utils.spec.ts (new): Covers ensureHttps, host detection, and normalization for Azure/generic providers, including query handling—guards against regressions in URL shaping.
openai-responses-adapter.ts (new): New adapter to decide Responses eligibility, build request bodies from Anthropic-style messages, stream Responses events into ApiStream, normalize usage, and run non-streaming completions. Enables Responses API integration without touching core handler logic.
openai-responses-mode.spec.ts (new): Tests OpenAiHandler decisions for streaming/non-streaming, ensuring Responses mode only triggers when configured, uses the correct client, and propagates store flag.
base-openai-compatible-provider.ts: Introduced Responses support for compatible providers—stores capability/mode flags, decides per-call whether Responses is required, builds Responses request bodies, and streams/completes via the new adapter while preserving legacy completions paths.
openai.ts: Major wiring for Responses API: normalizes base URLs, creates a dedicated Responses client without Azure api-version, computes effective mode (completions vs responses), routes to Responses adapter when needed (including store/max-tokens handling), and keeps Azure-specific query params off Responses calls. Maintains legacy completions behavior otherwise.
OpenAICompatible.tsx: Settings UI now lets users choose API style (Completions vs Responses), toggle Responses store, and auto-disables Azure api-version when Responses is selected; Azure api-version input is gated to Completions mode. Exposes the new settings to match backend expectations.

Screenshots

before	after

|

|

|

How to Test

Add a recent OpenAI model like GPT-5.1-codex-max. It wouldn't work without this pull request!
Set the provider api style to responses and have fun using the most recent LLMs!

- Added support for OpenAI Responses API in OpenAiHandler, including handling of base URLs and API modes. - Introduced utility functions for URL normalization and Azure host detection. - Created a new module for handling Responses API requests and streaming responses. - Updated OpenAI settings UI to allow users to select between Completions and Responses API modes. - Added tests for URL utilities and Responses API functionality. - Enhanced error handling for OpenAI API interactions.

changeset-bot · 2025-12-13T23:36:21Z

🦋 Changeset detected

Latest commit: 8fb8854

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	Major

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

mateuszhoffman · 2026-01-01T18:39:09Z

That would be awesome to have! Can anyone can approve this?

pedrommone · 2026-01-04T21:53:59Z

I'm thinking move away from Cursor just because their API is not fully compatible with Azure models. We need that since we rely on Azure.

Is there a way to get this PR approved? We can move ~10 licenses to Kilo due that.

kevinvandijk · 2026-01-14T15:05:09Z

Thank you! We'll review this soon. @chrarnoldus could you review this since you're more up to date on this than I am?

chrarnoldus · 2026-01-14T15:11:37Z

I think it should be made a separate provider "OpenAI Responses Compatible" or some such.

If the goal is purely to use Azure, we could also figure out why the "OpenAI" (native) provider with a custom base url doesn't work.

qbiecom · 2026-01-15T21:05:05Z

First major difference is that the API key is in a different header. Example from their docs:

curl -X POST "https://<REDACTED>-uksouth.cognitiveservices.azure.com/openai/responses?api-version=2025-04-01-preview" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $AZURE_API_KEY" \
    -d '{
        "messages": [
            {
                "role": "user",
                "content": "I am going to Paris, what should I see?"
            }
        ],
        "max_completion_tokens": 16384,
        "model": "gpt-5.1"
    }'

That's in addition to some of the newer models that only support the responses API (5.1/5.2 codex for example).

qbiecom · 2026-01-20T17:55:49Z

Updating my previous comment - Azure works fine using OpenAI compatible for models that support chat completion, as long as you tick the 'Use Azure' option in Kilo.

It's just that they restrict some of their models to responses API only now.

kommerzienrat · 2026-01-20T19:48:16Z

Yes, that's why I took on this issue in the first place. And since the responses API is the future, we should include it anyway.

chrarnoldus · 2026-01-20T21:05:35Z

We can accept it as a separate provider from Chat Completions.

mateuszhoffman · 2026-01-21T10:04:11Z

So... will it be integrated?

qbiecom · 2026-01-21T10:49:55Z

In which case it'll probably be better to close this PR and then create a new one - clone the existing provider and update it for responses, and then name them 'OpenAI Compatible (Chat Completions)' and 'OpenAI Compatible (Responses)'?

chrarnoldus · 2026-01-23T19:42:02Z

Closing in favor of #5330, but thanks for the effort @kommerzienrat

kommerzienrat added 2 commits December 13, 2025 23:58

fix: Update translations for OpenAI Responses store settings

5a5060e

kommerzienrat and others added 3 commits December 14, 2025 00:42

feat: Add major update for OpenAI Responses API support

540d3d1

Merge branch 'main' into openai-responses-api

0e8f81f

Merge branch 'main' into openai-responses-api

2426f27

kevinvandijk self-requested a review December 15, 2025 13:31

kevinvandijk self-assigned this Dec 15, 2025

kommerzienrat added 3 commits December 16, 2025 09:32

Merge branch 'main' into openai-responses-api

23f6bdd

Merge branch 'main' into openai-responses-api

629efc5

Merge branch 'main' into openai-responses-api

370727d

Merge branch 'main' into openai-responses-api

b2a1493

kommerzienrat changed the title ~~OpenAI responses api~~ feat: OpenAI responses api Jan 13, 2026

Merge branch 'main' into openai-responses-api

8fb8854

kevinvandijk requested a review from chrarnoldus January 14, 2026 15:05

chrarnoldus closed this Jan 23, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: OpenAI responses api #4442

feat: OpenAI responses api #4442

Uh oh!

kommerzienrat commented Dec 13, 2025 •

edited

Loading

Uh oh!

changeset-bot bot commented Dec 13, 2025 •

edited

Loading

Uh oh!

mateuszhoffman commented Jan 1, 2026

Uh oh!

pedrommone commented Jan 4, 2026

Uh oh!

kevinvandijk commented Jan 14, 2026

Uh oh!

chrarnoldus commented Jan 14, 2026 •

edited

Loading

Uh oh!

qbiecom commented Jan 15, 2026 •

edited

Loading

Uh oh!

qbiecom commented Jan 20, 2026 •

edited

Loading

Uh oh!

kommerzienrat commented Jan 20, 2026

Uh oh!

chrarnoldus commented Jan 20, 2026

Uh oh!

mateuszhoffman commented Jan 21, 2026

Uh oh!

qbiecom commented Jan 21, 2026

Uh oh!

chrarnoldus commented Jan 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

feat: OpenAI responses api #4442

feat: OpenAI responses api #4442

Uh oh!

Conversation

kommerzienrat commented Dec 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Context

Implementation

Screenshots

How to Test

Uh oh!

changeset-bot bot commented Dec 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

mateuszhoffman commented Jan 1, 2026

Uh oh!

pedrommone commented Jan 4, 2026

Uh oh!

kevinvandijk commented Jan 14, 2026

Uh oh!

chrarnoldus commented Jan 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

qbiecom commented Jan 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

qbiecom commented Jan 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kommerzienrat commented Jan 20, 2026

Uh oh!

chrarnoldus commented Jan 20, 2026

Uh oh!

mateuszhoffman commented Jan 21, 2026

Uh oh!

qbiecom commented Jan 21, 2026

Uh oh!

chrarnoldus commented Jan 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

kommerzienrat commented Dec 13, 2025 •

edited

Loading

changeset-bot bot commented Dec 13, 2025 •

edited

Loading

chrarnoldus commented Jan 14, 2026 •

edited

Loading

qbiecom commented Jan 15, 2026 •

edited

Loading

qbiecom commented Jan 20, 2026 •

edited

Loading