You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Sep 23, 2026. It is now read-only.
Repository navigation
This repository was archived by the owner on Sep 23, 2026. It is now read-only.
Please make your kimi code api work as OpenAI-compatible API #2208
Supporting an OpenAI-compatible /v1 endpoint would make this much easier to use from Cursor and other coding tools, because users could configure it with the same base_url + API key shape they already use elsewhere.
Then Cursor / SDK clients can keep one OpenAI-compatible integration path instead of a provider-specific plugin.
I have a small examples repo that shows the base_url pattern for Cursor, Claude Code, Codex, SDKs, and curl; it may be useful as a reference for the docs if Kimi exposes an OpenAI-compatible endpoint:
From the screenshots, the request is reaching Kimi but is being rejected by that specific endpoint.
Cursor is configured with:
https://api.kimi.com/coding/v1
and the Postman request goes to:
https://api.kimi.com/coding/v1/chat/completions
The response is 403 Forbidden with:
{
"error": {
"message": "Kimi For Coding is currently only available for Coding Agents such as Kimi CLI, Claude Code, Roo Code, Kilo Code, etc.",
"type": "access_terminated_error"
}
}
So this does not look like a JSON/body-format problem. It looks like /coding/v1 is an access-restricted product endpoint, not a general OpenAI-compatible endpoint for arbitrary clients.
For Cursor/OpenAI-compatible usage, the docs probably need to clarify one of these paths:
If Kimi exposes a general OpenAI-compatible API, document that separate base URL, supported model names, and whether Cursor is allowed.
If https://api.kimi.com/coding/v1 is intentionally limited to Kimi CLI / Claude Code / Roo Code / Kilo Code, document that it is not expected to work from Cursor’s generic OpenAI override.
If Cursor should be supported, the server-side allowlist/client detection is probably what needs changing, because the request shape already reaches /chat/completions and fails at access control.
Activity
haocn-ops commentedon May 12, 2026
Supporting an OpenAI-compatible
/v1endpoint would make this much easier to use from Cursor and other coding tools, because users could configure it with the samebase_url+ API key shape they already use elsewhere.A common config shape is:
Then Cursor / SDK clients can keep one OpenAI-compatible integration path instead of a provider-specific plugin.
I have a small examples repo that shows the base_url pattern for Cursor, Claude Code, Codex, SDKs, and curl; it may be useful as a reference for the docs if Kimi exposes an OpenAI-compatible endpoint:
https://github.com/apinode-pro/openai-compatible-ai-gateway-examples
janeza2 commentedon May 12, 2026
here's what i found when i trying to use in cursor
then i tried it on postman so i got this
haocn-ops commentedon May 26, 2026
From the screenshots, the request is reaching Kimi but is being rejected by that specific endpoint.
Cursor is configured with:
and the Postman request goes to:
The response is
403 Forbiddenwith:{ "error": { "message": "Kimi For Coding is currently only available for Coding Agents such as Kimi CLI, Claude Code, Roo Code, Kilo Code, etc.", "type": "access_terminated_error" } }So this does not look like a JSON/body-format problem. It looks like
/coding/v1is an access-restricted product endpoint, not a general OpenAI-compatible endpoint for arbitrary clients.For Cursor/OpenAI-compatible usage, the docs probably need to clarify one of these paths:
https://api.kimi.com/coding/v1is intentionally limited to Kimi CLI / Claude Code / Roo Code / Kilo Code, document that it is not expected to work from Cursor’s generic OpenAI override./chat/completionsand fails at access control.janeza2 commentedon May 31, 2026
the reason that i show you Postman request: https://api.kimi.com/coding/v1/chat/completions
because I want to prove that kimi code doesn't support OpenAI-compatible right now
this open issue is basically feature request to kimi team