#Trying to switch to ollama/kimi-k2.6

1 messages · Page 1 of 1 (latest)

fierce isle
#

I'm trying to switch to the new ollama/kimi-k2.6 could but keeps failing.. I'm on 2026.4.20 I see it in the /model list .. what am I missing?

peak lanceBOT
inner yachtBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> âś… Mark Solution

lunar lakeBOT
#

Let me verify the docs and source first so I don’t hand-wave this.

#

I think you’re probably mixing provider refs.

kimi-k2.6 is documented as moonshot/kimi-k2.6, not ollama/kimi-k2.6.

For Ollama, the docs currently show cloud refs like ollama/kimi-k2.5:cloud, and those only work if:

  1. you’re using Cloud only with baseUrl: https://ollama.com + a real OLLAMA_API_KEY, or
  2. you’re using Cloud + Local and have run ollama signin on that Ollama host.

Also, important, if your Ollama URL ends with /v1, that’s wrong for OpenClaw. Use the native Ollama URL like http://host:11434 (no /v1).

So the likely misses are:

  • wrong model ref (moonshot/kimi-k2.6 vs ollama/...)
  • Ollama cloud not actually enabled/authenticated
  • Ollama base URL set to /v1
  • seeing it in /model list but your current provider config still can’t serve it
fierce isle
#

Current: ollama/kimi-k2.5:cloud
Default: ollama/kimi-k2.5:cloud
Agent: main
Auth file: ~/.openclaw/agents/main/agent/auth-profiles.json

[lmstudio] endpoint: http://192.168.7.138:1234/v1 api: openai-completions auth: lm...io (models.json: ~/.openclaw/agents/main/agent/models.json)
• lmstudio/qwen/qwen3.5-35b-a3b (Qwen Local)
• lmstudio/infrastructure (Infrastructure)

[ollama] endpoint: http://127.0.0.1:11434 (http://127.0.0.1:11434/) api: ollama auth: ollama:default=ol...al (next, lastGood) (auth-profiles.json: ~/.openclaw/agents/main/agent/auth-profiles.json)
• ollama/kimi-k2.5:cloud (Kimi Cloud)
• ollama/quick (Quick)
• ollama/sentinel (Sentinel)
• ollama/hermes (Hermes)
• ollama/forge (Forge)
• ollama/scout (Scout)
• ollama/scribe (Scribe)
• ollama/shopkeeper (ShopKeeper)
• ollama/bonanzakeeper (BonanzaKeeper)
• ollama/ebaykeeper (EbayKeeper)
• ollama/gemini-3-flash-preview:cloud (Gemini Flash Cloud)
• ollama/nemotron-3-super:cloud (Nemotron Super Cloud)
• ollama/kimi-k2.6

[google] endpoint: default auth: google:default=AIzaSyDE...46AvdfL8 (next) (auth-profiles.json: ~/.openclaw/agents/main/agent/auth-profiles.json)
• google/gemini-3.1-flash:free (Gemini Flash)

lunar lakeBOT
#

That paste helps a lot.

Your Ollama setup looks basically correct now:

  • native Ollama API, not /v1
  • model is being discovered
  • auth/provider are loaded

The main thing I notice is this:

  • you have ollama/kimi-k2.6
  • you do not have ollama/kimi-k2.6:cloud

So if you’re trying to switch to the :cloud ref, that would fail. For K2.6, use the exact ref that shows up:

/model ollama/kimi-k2.6
#

If you want to change the default too, not just this chat session:

openclaw models set ollama/kimi-k2.6

Then verify with:

openclaw models status

One more gotcha, if a run is already active, /model can stay pending until the next clean retry point or next message.

So I’d try, in order:

  1. /model ollama/kimi-k2.6
  2. send a fresh message after switching
fierce isle
#

[4/22/2026 9:33 AM] TaskMaster2 (.26): âś… Model changed to ollama/kimi-k2.6

This model will be used for your next message.
[4/22/2026 9:33 AM] Kloz: /new
[4/22/2026 9:34 AM] TaskMaster2 (.26): ⚠️ Something went wrong while processing your request. Please try again, or use /new to start a fresh session... then is switch model to default

lunar lakeBOT
#

I think I see the behavior now, let me verify the /new bit so I don’t mislead you.

#

Yep, that explains it.

/model is session-scoped, and bare /new starts a fresh session, so it falls back to the default model again. So this part is expected:

  • /model ollama/kimi-k2.6 → changes the current session
  • /new → creates a new session, uses default again (ollama/kimi-k2.5:cloud)

If you want a fresh chat on K2.6, do either:

/new ollama/kimi-k2.6

or:

# 
/new
/model ollama/kimi-k2.6

If you want all new sessions to start on K2.6, change the default:

openclaw models set ollama/kimi-k2.6

So, good news, the switch itself is working. The reset after /new is just session behavior.

The ⚠️ Something went wrong... part is probably a separate issue. If you try:

fierce isle
#

I typed the /new ollama/kimi-k2.6 and it just went back to kimi 2.5

lunar lakeBOT
#

Let me check whether /new <model> is actually wired the way the docs imply on Discord, because that sounds like a real bug, not you.

#

Yeah, that sounds like a real bug, not you.

I checked the docs/source, and /new <model> is supposed to work. But from your repro, the Discord native /new path is ignoring the model arg and starting a fresh session on the default anyway.

So for now the workarounds are:

  1. Fresh session, then switch it

    /new
/model ollama/kimi-k2.6

  2. Make K2.6 the default

    openclaw models set ollama/kimi-k2.6
#

Given what you pasted, I’d treat this as a bug in 2026.4.20.

If you can, please file it here with your exact repro:
https://github.com/openclaw/openclaw/issues

Good repro steps:

  1. /model ollama/kimi-k2.6 succeeds
  2. /new ollama/kimi-k2.6 starts a fresh session
  3. session comes back on default ollama/kimi-k2.5:cloud

I’m glad you pushed on it, this looks like a legit regression.

fierce isle
#

Failed to set model: GatewayRequestError: model not allowed: ollama/kimi-k2.6:cloud

#

Failed to set model: GatewayRequestError: model not allowed: ollama/kimi-k2.6

lunar lakeBOT
#

I’ve got the actual failure now, let me verify the allowlist bit so I point you at the right config key.

#

Got it, this is the real blocker.

ollama/kimi-k2.6 is being discovered by the provider, but it is not in your model allowlist.

OpenClaw treats agents.defaults.models as the allowlist for /model and session overrides, so you can end up in this exact state:

  • shows up in provider status
  • still fails with model not allowed

So the fix is one of these:

  1. Add it to agents.defaults.models
  2. Remove/clear agents.defaults.models if you do not want a strict model allowlist
  3. Re-run the model section in configure:
    openclaw configure --section model
#

Useful check:

openclaw config get agents.defaults.models

If you edit manually, you want ollama/kimi-k2.6 included there, for example:

{
  agents: {
    defaults: {
      models: {
        "ollama/kimi-k2.5:cloud": { alias: "Kimi Cloud" },
        "ollama/kimi-k2.6": { alias: "Kimi K2.6" }
      }
fierce isle
#

{
"google/gemini-3.1-flash:free": {
"alias": "Gemini Flash"
},
"lmstudio/infrastructure": {
"alias": "Infrastructure"
},
"lmstudio/qwen/qwen3.5-35b-a3b": {
"alias": "Qwen Local"
},
"ollama/bonanzakeeper": {
"alias": "BonanzaKeeper"
},
"ollama/ebaykeeper": {
"alias": "EbayKeeper"
},
"ollama/forge": {
"alias": "Forge"
},
"ollama/gemini-3-flash-preview:cloud": {
"alias": "Gemini Flash Cloud"
},
"ollama/hermes": {
"alias": "Hermes"
},
"ollama/kimi-k2.5:cloud": {
"alias": "Kimi Cloud"
},
"ollama/nemotron-3-super:cloud": {
"alias": "Nemotron Super Cloud"
},
"ollama/quick": {
"alias": "Quick"
},
"ollama/scout": {
"alias": "Scout"
},
"ollama/scribe": {
"alias": "Scribe"
},
"ollama/sentinel": {
"alias": "Sentinel"
},
"ollama/shopkeeper": {
"alias": "ShopKeeper"
}
}

lunar lakeBOT
#

Yep, that confirms it.

Your allowlist is missing:

"ollama/kimi-k2.6": { "alias": "Kimi K2.6" }

So add that under agents.defaults.models, then try the switch again.

Minimal diff:

{
  agents: {
    defaults: {
# 
      models: {
        // existing entries...
        "ollama/kimi-k2.6": { alias: "Kimi K2.6" }
      }
    }
  }
}

Then retry:

/model ollama/kimi-k2.6

or make it default: