Openclaw And Koboldcpp connection

[FlyToha]

Hi! I unsuccessfully trying to setup openclaw + coboldcpp on the pc.
What i did:
-Start koboldcpp on main pc at localhost and default port 5001. Qwen3.5-35b model selected.
-complete the basic setup Openclaw on the virtual machine
-from the vm check curl http://10.0.2.2:5001/v1/models - ok, i have got correct response with correct model name
-from the vm i joined to kobold lite thats meaning the koboldcpp 100% available from the vm
-edit openclaw config, added there this lines:

"models": {
"node": "merge",
"providers": {
"koboldcpp": {
"baseUrl": "http://10.0.2.2:5001/v1",
"apiKey": "123",
"api": "openai-responses",
"headers": {
"X-Proxy-Region": "OPENCLAN_REDACTED"
},
"authHeader": false,
"models": [
{
"id": "koboldcpp/Qwen3.5-35B-A3B-Uncensored-HaukauCS-Aggressive-Q6_K",
"name": "Qwen3.5-35B-A3B",
"api": "openai-responses",
"reasoning": false,
"input": [
"text"
],
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"contextWindow": 128000,
"maxTokens": 32000
}

But when i try to ask some thing the bot via webui chat i have got the follow error: 404 status code (no body). Openclaw logs here:

20:30:43+00:00 warn agent/embedded {"subsystem":"agent/embedded"} {
"event": "embedded_run_agent_end",
"tags": ["error_handling", "lifecycle", "agent_end", "assistant_error"],
"runId": "0e9946811-3ayb-4db9-b580-5f7e46658ac4",
"isError": true,
"error": "404 status code (no body)",
"failoverReason": null,
"model": "koboldcpp/Qwen3.5-35B-A3B-Uncensored-HaukauCS-Aggressive-Q6_K",
"provider": "koboldcpp",
"rawErrorPreview": "404 status code (no body)",
"rawErrorHash": "sha256:b02359c4c991"
}
embedded run agent end
Log tail truncated (increase --max-bytes).

Can someone help me to setup connection between OpenClaw on a virtual machine and host pc with Koboldcpp? Of course i've tried it with llms, but that bots didn't help in this case.

[gustrd]

For me it works like this:

"providers": {
      "kcpp": {
        "baseUrl": "http://127.0.0.1:5001/v1",
        "apiKey": "-",
        "api": "openai-completions",
        "models": [
          {
            "id": "kcpp",
            "name": "kcpp (Custom Provider)",
            "reasoning": true,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 16384,
            "maxTokens": 2048
          }
        ]
      },

Try "openai-completions" instead of "openai-responses". I also don't need ""X-Proxy-Region".

[LostRuins]

Yeah the issue is the line "api": "openai-responses", change that to "api": "openai-completions",

In general the responses API sucks and shouldnt be used

[FlyToha]

Wow guys, it's works now! I've tried "openai-completions" and its start working immediately. Thank you both so much!

[gustrd]

Good!

There is a common issue, because openclaw always uses streaming responses, and sometime a tool call at the middle of a message is missed.

I'm using a middleware to avoid it. @LostRuins , do you have a better idea?

[LostRuins]

I don't have a good solution. I'm planning to eventually add basic /responses endpoint support but it'll definitely be more primitive than chat completions.

[LostRuins]

@gustrd i added 5ad9e3e

[gustrd]

Thanks @LostRuins! Please give me some days to try and I will get back.

[jingchang0623-crypto]

[FlyToha]

Lol, i spent 2 days to try solve my problem and any llms told me exactly same useless answers like this one. But when i asked here the first human just point me the true solution immediately. So i think the riot of the machines postponed, they are still too dumb :D

[LostRuins]

deleted the useless bot comment, spam is bad