#Openclaw Not using api-keys on cloud modles

1 messages · Page 1 of 1 (latest)

hexed knot
#

Hey there, my openclaw instance has 3 different cloud llms providers and 2 local ollama LLM nodes.
It only uses the ollama cloud llms.
All others are not working, I guess the tomorrow for local models is set to short but I can't find the setting.
Also my cloud models from Openrouter and groq aren't used and fail instantly.
Of I check the dashboards I see a lot of failed 404 requests

vast turretBOT
thin lindenBOT
#

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

vast turretBOT
hexed knot
#

I'm running openclaw on a Ubuntu 24.04 VM on my proxmox cluster.
It has full access to the system and the ollama cloud models work for my agent but all others fail, so I expect wrong setups by openclaw.
How can I find the correct config and fix them

dreamy totemBOT
#

First, what does openclaw --version show?

From the docs, this does not sound like a hidden “prefer local models” setting. OpenClaw picks models in this order:

  1. agents.defaults.model.primary
  2. agents.defaults.model.fallbacks in order

So if Ollama is always being used, the usual cause is that your active/default model is still ollama/..., or your cloud providers are misconfigured and failing immediately.

The two big things I’d check next are:

  • Wrong default model / fallback order
  • Wrong provider endpoint or model ref, especially if you’re seeing 404s

A few important gotchas from the docs:

  • OpenRouter should be using the OpenRouter API route, https://openrouter.ai/api/v1
#
  • Groq uses https://api.groq.com/openai/v1
  • Ollama should use the native Ollama API URL, like http://host:11434, not .../v1

Please paste these outputs, with keys/tokens redacted:

openclaw --version
openclaw models status
openclaw models status --probe
openclaw config get agents.defaults.model
openclaw config get models.providers

If you can, also reproduce one failed cloud call and paste the matching lines from:

#

Yep, I’d debug this in this order.

There is not a hidden “prefer local models” knob. OpenClaw uses:

  1. agents.defaults.model.primary
  2. agents.defaults.model.fallbacks

So if Ollama always wins, either the active/default model is still ollama/..., or the cloud providers are failing immediately.

1. Find the active config

OpenClaw config is JSON5, usually at:

~/.openclaw/openclaw.json

Verify the gateway is actually using that config:

# 
openclaw --version
openclaw health --verbose
openclaw gateway status

2. Check what model/auth the gateway sees

openclaw status
openclaw models status
openclaw models status --probe
openclaw config get agents.defaults.model
openclaw config get models.providers

What I’d expect to catch there:

#
  • agents.defaults.model.primary still points at ollama/...
  • missing auth for OpenRouter or Groq
  • wrong baseUrl
  • wrong model refs for the provider

3. On Ubuntu, check daemon env

This is a very common gotcha. If the gateway runs as a service, keys in your interactive shell often are not visible to it.

Docs say env comes from the process env, .env files, then config env, without overriding existing values. The safest place for provider keys on a VM is usually:

~/.openclaw/.env

For example:

OPENROUTER_API_KEY=...
# 
GROQ_API_KEY=...

Then restart the gateway.

4. Verify provider-specific setup

A 404 usually means the provider endpoint/path is wrong.

  • OpenRouter should go to https://openrouter.ai/api/v1
  • Groq should go to https://api.groq.com/openai/v1
  • Ollama should use the native Ollama URL like http://host:11434, not /v1

5. After auth is fixed, switch the default model

openclaw models list --provider openrouter
# 
openclaw models list --provider groq
openclaw models set <provider/model>

If you paste these outputs, with secrets redacted, I can usually spot the bad entry quickly:

openclaw --version
openclaw models status
openclaw models status --probe
openclaw config get agents.defaults.model
openclaw config get models.providers

Docs:

hexed knot
#

Can't I add screenshots in here?

dreamy totemBOT
#

Yep, screenshots are fine here.

Best ones to share would be:

  • openclaw models status
  • openclaw models status --probe
  • the relevant part of models.providers / agents.defaults.model
  • the 404 error in the dashboard or logs

Just redact API keys, tokens, and any private hostnames first.

hexed knot
#

I don't have permission to post screenshots or rather they are fully disabled here, I can't attach any data besides texting

dreamy totemBOT
#

No worries, text is better here anyway.

Please paste these, each in its own code block:

openclaw --version
openclaw models status
openclaw models status --probe
openclaw config get agents.defaults.model
openclaw config get models.providers

If models.providers is huge, just paste the openrouter, groq, and ollama parts.

Then trigger one failed cloud request and paste the matching log lines from: