KIMI cannot work

[luoyeguigenno1]

here is the api configuration (kimi coding api key), connection is sucessfull, but the chat cannot work.

i have tried all combinations, the connection is always successful, but the chat cannot work.

changing the model to kimi-k2 does not work.

How can I fix it?

[hqhq1025]

Hi, this issue may be related to the API configuration logic in older versions.

We have recently fixed several API-related issues in the latest release. Please try the new version here:

https://github.com/OpenCoworkAI/open-cowork/releases/tag/v3.3.0-beta2

This version includes fixes for the API handling and model configuration.
Could you update to this version and test again to see if the problem is resolved?

If it still doesn’t work, please share your configuration (base URL, model name, provider type), and we can help further investigate.

[luoyeguigenno1]

Hi, this issue may be related to the API configuration logic in older versions.

We have recently fixed several API-related issues in the latest release. Please try the new version here:

https://github.com/OpenCoworkAI/open-cowork/releases/tag/v3.3.0-beta2

This version includes fixes for the API handling and model configuration. Could you update to this version and test again to see if the problem is resolved?

If it still doesn’t work, please share your configuration (base URL, model name, provider type), and we can help further investigate.

Hi, thanks greatly for your update. It still does not work in open cowork, but openclaw can use this api with the same url and api key. here is the output of open cowork, for which i have tried differnent combinations

here is the o output of openclaw, which indicates the api can work

[hqhq1025]

Hi, you’re right — from your screenshot the connection probe itself did not pass. The error empty_probe_response means Open Cowork sent the probe request, but the provider returned no usable text response, so chat could not start afterward.

Please try creating a new config set and test again with a clean setup, since this is still a beta build and there may be some config-state bugs when switching providers or protocols.

For Kimi Coding, please try:

Create a new config set
Protocol: Anthropic
Base URL: https://api.kimi.com/coding
Model: use the exact coding model ID from Moonshot/Kimi docs for that endpoint
If it still fails, please send us:

the exact model ID
whether Live request verification was enabled
the full error text shown in the settings panel
We’ll try to reproduce this on our side. If we can reproduce it or fully confirm the root cause, we’ll update you here once we have a proper fix or a confirmed configuration workaround.

[luoyeguigenno1]

Hi, you’re right — from your screenshot the connection probe itself did not pass. The error empty_probe_response means Open Cowork sent the probe request, but the provider returned no usable text response, so chat could not start afterward.

Please try creating a new config set and test again with a clean setup, since this is still a beta build and there may be some config-state bugs when switching providers or protocols.

For Kimi Coding, please try:

Create a new config set Protocol: Anthropic Base URL: https://api.kimi.com/coding Model: use the exact coding model ID from Moonshot/Kimi docs for that endpoint If it still fails, please send us:

the exact model ID whether Live request verification was enabled the full error text shown in the settings panel We’ll try to reproduce this on our side. If we can reproduce it or fully confirm the root cause, we’ll update you here once we have a proper fix or a confirmed configuration workaround.

hi thanks for your prompt response. but it does not work. i can even share my api key if needed, but safety maybe an issue.

here is the api key configuration

here is the setting page. i have tried different models.

[github-actions]

Thanks for reporting this. I checked the current codebase and Kimi Coding is wired through the generic custom-provider flow, not a separate Kimi-specific adapter. In the repo, the expected setup is Custom provider + Anthropic protocol + https://api.kimi.com/coding as the base URL (src/shared/api-provider-guidance.ts:75, src/shared/api-provider-guidance.ts:80, src/main/claude/pi-model-resolution.ts:42).

I also verified one inconsistency that may matter here: the README still lists Kimi as kimi-k2 (readme.md:160), but the in-app Kimi guidance and custom model presets use kimi-k2-thinking (src/shared/api-provider-guidance.ts:81, src/shared/api-model-presets.ts:103). Both the settings probe and the normal chat runtime use the same routing/protocol resolution path (src/main/claude/claude-sdk-one-shot.ts:106, src/main/claude/claude-sdk-one-shot.ts:298, src/main/claude/agent-runner.ts:1271), so the exact protocol and model ID are the first things to confirm.

Relevant code:

• src/shared/api-provider-guidance.ts:75 - Kimi Coding is mapped to custom + anthropic
• src/shared/api-provider-guidance.ts:80 - recommended base URL is https://api.kimi.com/coding
• src/shared/api-provider-guidance.ts:81 - example Kimi model is kimi-k2-thinking
• src/main/claude/pi-model-resolution.ts:42 - custom providers default to Anthropic unless openai or gemini is explicitly selected
• src/main/claude/claude-sdk-one-shot.ts:106 - connection probe builds its config from the selected provider/protocol/base URL
• src/main/claude/claude-sdk-one-shot.ts:298 - diagnostics/test connection uses the one-shot probe
• src/main/claude/agent-runner.ts:1271 - normal chat runtime resolves the provider/protocol again for actual sessions
• readme.md:160 - README currently shows kimi-k2, which does not match the in-app example

Need more info:

• What Open Cowork version are you using?
• In Settings, what are the exact Provider, Protocol, Base URL, and Model values? Please redact the API key.
• Does the failure happen in Test connection, in normal chat, or both?
• Please paste the full error text as text instead of screenshots.

---

Open Cowork Bot

[hqhq1025]

Closing as stale — no activity for over a month. If this is still an issue, feel free to reopen or file a new issue. Thanks!