ollama model discovery | Friends of the Crustacean 🦞🤝 | Page 1

vale lichen Apr 13, 2026, 2:08 PM

#

As per the docs "When you set OLLAMA_API_KEY (or an auth profile) and do not define models.providers.ollama, OpenClaw discovers models from the local Ollama instance at http://127.0.0.1:11434." Why don't I see all the models in openclaw?

`$ollama list
NAME ID SIZE MODIFIED
qwen3.5:397b-cloud a7bf6f7891c3 - 3 minutes ago
minimax-m2.7:cloud 06daa293c105 - 5 minutes ago
glm-5.1:cloud 59472abf9d0a - 2 days ago

$ openclaw models list
🦞 OpenClaw 2026.4.11 (unknown) — I'll refactor your busywork like it owes me money.

Model Input Ctx Local Auth Tags
ollama/glm-5.1:cloud text 198k yes yes default,configured
nano-gpt/qwen3.5-122b-a10b:thinking text+image 254k no yes image
nano-gpt/Qwen3.5-27B-Derestricted text+image 256k no yes configured,alias:deres
nano-gpt.com/minimax/minimax-m2.7 text 200k no yes configured
nano-gpt/nvidia/nemotron-3-super-120b-a12b text 256k no yes configured,alias:fast
nano-gpt/nvidia/nemotron-3-super-120b-a... text 256k no yes configured
openrouter/nvidia/nemotron-3-super-120b... text 256k no yes configured
openrouter/nvidia/nemotron-3-nano-30b-a... text 250k no yes configured
node@f7619ca23724:~/.openclaw$ openclaw models list --all | grep ollama

node@f7619ca23724:~/.openclaw$ openclaw models list --all --provider ollama

🦞 OpenClaw 2026.4.11 (unknown) — If you're lost, run doctor; if you're brave, run prod; if you're wise, run tests.

No models found.`

thorn stumpBOT Apr 13, 2026, 2:08 PM

#

ancient flareBOT Apr 13, 2026, 2:08 PM

#

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

thorn stumpBOT Apr 13, 2026, 2:08 PM

#

topaz surgeBOT Apr 13, 2026, 2:11 PM

#

I’m checking the discovery/filtering path in code too, because that models list vs models list --provider ollama mismatch smells like a real bug, not just config.

#

openclaw models list and the doc snippet are talking about two slightly different paths.

What I think is happening:

openclaw models list shows configured models by default, not every discovered Ollama tag.
The implicit auto-discovery from the docs only applies when you do not have an explicit models.providers.ollama config.
If you ran onboarding/manual Ollama setup before, OpenClaw likely wrote an explicit models.providers.ollama block, and then later Ollama additions will not auto-appear.

Also, implicit discovery is hard-coded to the local Ollama API at http://127.0.0.1:11434, so if your ollama CLI is pointed at a different host, the lists can differ.

The weird part is this:

openclaw models list shows ollama/glm-5.1:cloud
but openclaw models list --all --provider ollama shows nothing

That looks inconsistent to me on 2026.4.11, and may be a CLI bug or a stale/manual Ollama provider config.

#

Can you check these two?

openclaw config get models.providers.ollama --json
curl http://127.0.0.1:11434/api/tags

If models.providers.ollama exists, that’s probably the reason. Then you have two options:

keep explicit config, and add the missing models under models.providers.ollama.models
remove that block and go back to implicit discovery

Relevant docs:

If you paste the models.providers.ollama output, I can tell you exactly which side you’re on.

vale lichen Apr 13, 2026, 2:20 PM

#

ok interestingly I did not do this manually, but checking openclaw.json I see that openclaw actually added a models.providers.ollama.models section and did populate the three models from 'ollama list' So I guess I do need to manually add that to the defaults.agents.models section?

topaz surgeBOT Apr 13, 2026, 2:20 PM

#

Yep, that’s the missing piece.

These are two different layers:

models.providers.ollama.models = provider catalog, meaning “Ollama has these models”
agents.defaults.models = allowlist/catalog for selection in OpenClaw

So, no, you do not need to duplicate them there just to make Ollama work.

But if agents.defaults.models is set, then only models listed there are treated as selectable/allowed for /model and the normal openclaw models list view.

So your options are:

Add the missing ollama/... refs to agents.defaults.models if you want them selectable
Remove/clear agents.defaults.models if you do not want an allowlist at all

And separately, keep your actual default in:

#

agents.defaults.model.primary

So in your case, yes, if you want those other Ollama models to show up as normal selectable models, add them to agents.defaults.models or remove that allowlist.

Docs:

One small note: openclaw models list --all --provider ollama returning nothing still looks a bit wrong to me on 2026.4.11. If that keeps reproducing after the allowlist fix, that’s worth a GitHub issue: https://github.com/openclaw/openclaw/issues

vale lichen Apr 13, 2026, 2:29 PM

#

ok I deleted all the agents.defaults.models and restarted, and still don't see them

$ openclaw config get agents.defaults

🦞 OpenClaw 2026.4.11 (unknown) — Give me a workspace and I'll give you fewer tabs, fewer toggles, and more oxygen.

{
"model": {
"primary": "ollama/glm-5.1:cloud"
},
"imageModel": {
"primary": "nano-gpt/qwen3.5-122b-a10b:thinking"
},
"workspace": "/home/node/.openclaw/workspace",
"memorySearch": {
"provider": "openai",
"remote": {
"baseUrl": "https://openrouter.ai/api/v1/",
"apiKey": "OPENCLAW_REDACTED"
},
"model": "nvidia/llama-nemotron-embed-vl-1b-v2:free"
},
"compaction": {
"mode": "safeguard"
},
"thinkingDefault": "high",
"maxConcurrent": 4,
"subagents": {
"maxConcurrent": 8
}
}
node@f7619ca23724:~/.openclaw$ openclaw models list

🦞 OpenClaw 2026.4.11 (unknown) — The lobster in your shell. 🦞

Model Input Ctx Local Auth Tags
ollama/glm-5.1:cloud text 198k yes yes default
nano-gpt/qwen3.5-122b-a10b:thinking text+image 254k no yes image

topaz surgeBOT Apr 13, 2026, 2:30 PM

#

I think we’ve narrowed it down: your plain models list output is now behaving normally, and the remaining problem is specifically the --all path. I’m checking whether that’s a known regression before I tell you to file it.

#

Yep, the plain openclaw models list output is now actually expected.

With no agents.defaults.models, that command shows the configured/default rows only, so in your case:

agents.defaults.model.primary → ollama/glm-5.1:cloud
agents.defaults.imageModel.primary → nano-gpt/qwen3.5-122b-a10b:thinking

That’s why you only see 2 rows.

So the remaining weirdness is just this part:

openclaw models list --all --provider ollama

That should be the command that shows the Ollama provider catalog, and since you already confirmed models.providers.ollama.models contains the 3 Ollama models, I’d call that unexpected on 2026.4.11.

#

Quick sanity check:

openclaw models set ollama/minimax-m2.7:cloud
openclaw models set ollama/qwen3.5:397b-cloud

If those work, then the provider catalog is fine and the bug is specifically in the listing path.

So my read is:

plain models list behavior: normal
--all --provider ollama returning nothing: likely bug/regression

If the direct models set ... works, I’d file it here with the outputs you already collected:
https://github.com/openclaw/openclaw/issues

#ollama model discovery