No responses back via local API call | Friends of the Crustacean 🦞🤝 | Page 1

drifting mason Apr 12, 2026, 8:26 PM

#

I have installed openclaw on a new Linux machine, it sees the ollama server and API sent after the question is asked. See system resources spike but no response back. Same for a windows machine.

All ports open, av exclusions. The only thing that works is when the cloud is an option. Have 32gb of ram. If asked locally on the ollama server I get a response back within 60 seconds or so.

Any help is appreciated.

limber plinthBOT Apr 12, 2026, 8:26 PM

#

wild ravineBOT Apr 12, 2026, 8:26 PM

#

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

wise wedgeBOT Apr 12, 2026, 8:28 PM

#

Let me check the current docs paths for the FAQ and Ollama guidance, then I’ll give you the tightest troubleshooting path.

#

Yep, this sounds more like a local-model compatibility/config issue than ports, especially since cloud works and plain Ollama responds.

First, can you paste these:

openclaw --version
openclaw status --all
openclaw models status

Then reproduce it once with:

openclaw logs --follow

and paste the last warning/error lines.

#

Two high-probability checks right away:

Do not use Ollama’s /v1 URL with OpenClaw.
Use http://host:11434, not http://host:11434/v1
If you configured models.providers.ollama manually, make sure it has:
```
api: "ollama"
```

I’m especially suspicious of one of these:

OpenAI-compatible /v1 mode instead of native Ollama API
a smaller/heavily quantized local model that works in plain Ollama, but chokes on OpenClaw’s larger agent prompt + tool schema
context/window pressure showing up only inside OpenClaw

Also, what exact Ollama model are you using?

Docs:

#

https://docs.openclaw.ai/help/faq
https://docs.openclaw.ai/providers/ollama
https://docs.openclaw.ai/gateway/local-models

If you want, paste your models.providers.ollama block too, with secrets redacted, and I’ll sanity-check it.

drifting mason Apr 12, 2026, 8:30 PM

#

Openclaw is on the newest version

No v1 at the end of the port is being used

Qwen2.5 chat version as cloud said it would work best.

wise wedgeBOT Apr 12, 2026, 8:31 PM

#

I’m going to check whether the docs call out any model-specific/local-model behavior here, then I’ll suggest the narrowest next test.

#

Got it. Then I’m a lot less suspicious of networking, and more suspicious of the local model itself under OpenClaw’s full agent prompt.

Important nuance: “Qwen cloud works well” does not automatically mean the local Ollama Qwen2.5 chat build will behave well in OpenClaw. Plain Ollama prompts are much lighter than an agent run.

Can you paste the exact output of:

openclaw --version
openclaw models status
openclaw infer model run --model ollama/<your-exact-model-id> --prompt "Reply with just OK." --json

Then in another terminal:

openclaw logs --follow

#

and reproduce one failed message.

What I’m trying to separate:

If openclaw infer model run also hangs/fails → it’s almost certainly the local model/backend choking on OpenClaw’s prompt shape/tool surface.
If infer model run works but Discord doesn’t reply → then we look at channel/run routing instead.

Also, as a quick A/B test, try one other local Ollama model just to rule out Qwen2.5 specifically. If another local model replies, that’s your smoking gun.

Relevant docs:
https://docs.openclaw.ai/providers/ollama
https://docs.openclaw.ai/gateway/local-models
https://docs.openclaw.ai/help/troubleshooting

Paste those outputs and I’ll narrow it down fast.

drifting mason Apr 12, 2026, 8:32 PM

#

I have also used gemma4 and qwen3.5 will all the same issues. I'll run the commands above and let you know what I see

wise wedgeBOT Apr 12, 2026, 8:32 PM

#

That’s actually useful, it makes a single bad model much less likely.

If it’s happening with gemma4, qwen2.5, and qwen3.5, on both Linux and Windows, I’d focus on one of these next:

a shared OpenClaw config/setup issue
the local backend handling the request, but choking on the full agent prompt/tool surface
a run failing silently and only showing up in logs

When you have it, please paste:

openclaw --version
openclaw models status
the result of openclaw infer model run ...
the last relevant lines from openclaw logs --follow

If you can, also paste your redacted models.providers.ollama block. That may be the fastest way to spot it.

drifting mason Apr 12, 2026, 8:59 PM

#

Version: 2026.4.9

Not sure what model status I'm looking at

Have to wait for the infer message. Getting to many requests

From follow commands just: warn bonjour error

Models.providors.ollama block no commands found

wise wedgeBOT Apr 12, 2026, 9:00 PM

#

I’m going to verify the exact config commands so I don’t send you a made-up key/path.

#

I’m checking whether that Bonjour warning matters here, so I don’t send you on a wild goose chase.

#

Perfect, that helps.

2026.4.9 is recent enough
bonjour warnings are probably unrelated here, that’s discovery/mDNS noise, not usually an Ollama reply blocker
for the config path, the command is:

openclaw config get models.providers.ollama --json

Also please run these exact ones:

openclaw models status
openclaw config get agents.defaults.model --json
openclaw doctor

#

A couple notes:

You typed providors, the path is providers
When I said “model status”, I meant the literal command:
```
openclaw models status
```

The most useful next paste would be:

output of openclaw models status
output of openclaw config get models.providers.ollama --json
output of openclaw config get agents.defaults.model --json
the full “too many requests” message, if it appears again

If infer is tripping rate limits, just wait for it to clear and run one test. That full error text matters a lot.

drifting mason Apr 12, 2026, 9:09 PM

#

Config path: gemma4:latest, qwen3.5:lastest, gemma4-8k:latest, gemma4-16k:latest, gemma4- 64k:latest

Model status: ollama:default=marker (ollama-local) how much is needed here?

Ran config get agents and nothing showed

wise wedgeBOT Apr 12, 2026, 9:10 PM

#

That helps. What you pasted means:

the models are being discovered
Ollama auth is present
but it still doesn’t tell us the resolved active model or whether the actual OpenClaw run succeeds

A couple clarifications:

ollama:default=marker (ollama-local) is just the auth marker, not the reply output
config get agents is too broad, use the exact path below

Please run these exact commands:

openclaw models status --plain
openclaw config get agents.defaults.model.primary

#No responses back via local API call