feat(providers/kimicode): add native Responses API support

[weselben]

TL;DR

The kimicode provider translated Responses API requests through chat completions, losing reasoning and usage shape. Kimi Code now serves the OpenAI Responses API natively at /responses. This PR forwards /v1/responses natively instead.

Files to review (3, +270 / -14):

File	Why
internal/providers/kimicode/kimicode.go (start here)	Native Responses/StreamResponses overrides plus request adaptation.
internal/providers/kimicode/kimicode_test.go	Unit tests: adaptation rules and native round-trips.
docs/providers/kimicode.mdx	Documents the native Responses path and its caveats.

Research

• Official Kimi Code docs list Codex as a supported third-party tool; Codex uses the OpenAI Responses API: https://www.kimi.com/code/docs/en/third-party-tools/codex
• Kimi platform docs list OpenAI Responses (/responses) as a supported protocol: https://platform.kimi.ai/docs/api/overview
• Live upstream probe (2026-09-08): non-streaming returns a complete response object with reasoning and usage; streaming returns standard Responses SSE events; store: true returns 400; previous_response_id returns 400.
• A https://api.kimi.ai/coding/v1 mirror answers with identical data. It is undocumented, and the official docs still list api.kimi.com/coding/v1. The base URL stays unchanged.

Reviewer notes

• State is adapted, not rejected. The upstream retains no responses, so store: true is rewritten to false and previous_response_id is dropped before forwarding (Postel's law). Both would otherwise fail upstream with a 400.
• Two adapter instances. The provider embeds openai.ChatCompatible for chat, models, embeddings, and passthrough, and holds an openai.CompatibleProvider for the native Responses transport. SetBaseURL updates both.
• Bearer headers are set explicitly. NewCompatibleProvider applies no SetHeaders default, unlike NewChatCompatible.

Tests

• go build ./... and go test ./... are green (run after make frontend for the dashboard-asset tests).
• New tests cover: request adaptation (store pin, previous_response_id drop, no caller mutation), a non-streaming native round-trip against a recorded upstream shape, and streaming passthrough including the [DONE] marker appended by providers.EnsureResponsesDone.

Links

• Context issue: feat(providers/kimicode): add native Responses API support weselben/GoModel#90

---

This PR description was generated with AI assistance.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 92.30769% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
internal/providers/kimicode/kimicode.go	92.30%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[weselben]

Review summary

Automated review of the draft. Findings: 1 yellow (fixed), 1 blue (fixed), 1 question (answered below). No red findings.

• 🟡 Stream-test wire assertion was vacuous — fixed in 7ea25c0.
• 🔵 Package comment pointed at a nonexistent responses.go — fixed in 7ea25c0.
• ❓ Upstream 400 behavior for store and previous_response_id — answered with probe evidence below.

Code change stays small and follows the existing chatgpt provider pattern. LGTM pending maintainer review.