ollama cloud issue | Friends of the Crustacean 🦞🤝 | Page 1

unreal knot Apr 13, 2026, 12:14 PM

#

I get the following error

08:11:26-04:00 [diagnostic] lane task error: lane=main durationMs=4152 error="FailoverError: Unknown model: ollama/glm-5.1:cloud. Ollama requires authentication to be registered "
08:11:26-04:00 [diagnostic] lane task error: lane=session:agent:claw-man:discord:channel:1481308516471013651 durationMs=4153 error="FailoverError: Unknown model: ollama/glm-5.1:cigure". See: https://docs.openclaw.ai/providers/ollama"
08:11:26-04:00 [model-fallback/decision] model fallback decision: decision=candidate_failed requested=ollama/glm-5.1:cloud candidate=ollama/glm-5.1:cloud reason=model_not_found ntered as a provider. Set OLLAMA_API_KEY="***" (any value works) or run "openclaw configure". See: https://docs.openclaw.ai/providers/ollama

But I do have OLLAMA_API_KEY set, and can chat with this model from command line using "ollama chat" How to fix?

scenic fableBOT Apr 13, 2026, 12:14 PM

#

dawn starBOT Apr 13, 2026, 12:14 PM

#

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

scenic fableBOT Apr 13, 2026, 12:14 PM

#

lean torrentBOT Apr 13, 2026, 12:16 PM

#

Let me verify the exact docs path and the Ollama provider notes, then I’ll pin down what’s actually going wrong.

#

I’m checking one more thing: whether OpenClaw is only missing the env var, or whether it also needs the cloud model registered explicitly.

#

This usually means OpenClaw doesn’t see Ollama as a configured provider, even if ollama chat works fine.

Most likely cause: OLLAMA_API_KEY is set in your shell, but the OpenClaw gateway/service was started from a different environment, so it never sees that var.

Fastest fix:

pnpm openclaw onboard

Then choose:

Ollama
Cloud + Local

That does 2 important things for you:

#

registers the Ollama provider for OpenClaw
adds the cloud model entries like ollama/glm-5.1:cloud

Then restart the gateway:

pnpm openclaw gateway restart

A couple notes:

ollama chat working does not prove the gateway sees OLLAMA_API_KEY
if you run OpenClaw via Docker/systemd/service, the env var must be set there, not only in .bashrc / terminal
one of your log lines shows ollama/glm-5.1:cigure, which looks suspiciously like a typo or bad fallback entry, so I’d also double-check the configured model string

If it still fails, paste these:

#

pnpm openclaw --version
pnpm openclaw models status
pnpm openclaw models list

Docs: https://docs.openclaw.ai/providers/ollama and https://docs.openclaw.ai/help/faq

unreal knot Apr 13, 2026, 12:18 PM

#

openclaw models status

🦞 OpenClaw 2026.4.11 (unknown) — Half butler, half debugger, full crustacean.

Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/main/agent
Default : nano-gpt.com/minimax/minimax-m2.7
Fallbacks (5) : ollama/glm-5.1:cloud, nano-gpt/nvidia/nemotron-3-super-120b-a12b:thinking, nano-gpt/nvidia/nemotron-3-super-120b-a12b, openrouter/nvidia/nemotron-3-super-120b-a12b:free, openrouter/nvidia/nemotron-3-nano-30b-a3b:free
Image model : nano-gpt/qwen3.5-122b-a10b:thinking
Image fallbacks (0): -
Aliases (2) : deres -> nano-gpt/Qwen3.5-27B-Derestricted, fast -> nano-gpt/nvidia/nemotron-3-super-120b-a12b
Configured models (7): nano-gpt/Qwen3.5-27B-Derestricted, nano-gpt.com/minimax/minimax-m2.7, ollama/glm-5.1:cloud, nano-gpt/nvidia/nemotron-3-super-120b-a12b, nano-gpt/nvidia/nemotron-3-super-120b-a12b:thinking, openrouter/nvidia/nemotron-3-super-120b-a12b:free, openrouter/nvidia/nemotron-3-nano-30b-a3b:free

Auth overview
Auth store : ~/.openclaw/agents/main/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (0): -

kilocode effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | kilocode:default=eyJhbGci...XWJrFZ7I
nano-gpt effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | nano-gpt:default=sk-nano-...0ba17fb8
nano-gpt.com effective=models.json:$N...EY | models.json=$N...EY | source=models.json: ~/.openclaw/agents/main/agent/models.json
ollama effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | ollama:default=marker(ollama-local) | env=3f3e1c9e...XcmnWAoD | source=env: OLLAMA_API_KEY
...

lean torrentBOT Apr 13, 2026, 12:21 PM

#

That helps a lot. I’m checking whether this is actually an agent-specific config mismatch now, because your main agent clearly does see OLLAMA_API_KEY.

#

Nice, this narrows it down a lot.

Your models status says main can see Ollama just fine:

source=env: OLLAMA_API_KEY
Configured models includes ollama/glm-5.1:cloud

So this is not a simple “missing env var” problem.

The big clue is in your log:

lane=session:agent:claw-man:discord:channel:...

That thread is being handled by agent claw-man, but the status you pasted is for agent main.

#

In OpenClaw, agents are isolated, so main and claw-man can have different:

auth profiles
models.json
session state

Also, Shell env: off is a red herring here. It does not mean the env var is missing, only that automatic shell-import is off.

What I’d do next

pnpm openclaw gateway restart
pnpm openclaw models status --agent claw-man
pnpm openclaw agents list --bindings

What I expect

#

If claw-man does not show ollama/glm-5.1:cloud or does not show Ollama auth, that’s the bug: the routed agent is missing the Ollama setup even though main has it.

If claw-man looks correct too, then I’d suspect stale session state. In that case, after restart, try a fresh thread/session (/new or /reset) because that weird ollama/glm-5.1:cigure line looks like stale or corrupted model state.

If you paste the output of:

pnpm openclaw models status --agent claw-man

I can tell you the exact fix from there.

Docs: https://docs.openclaw.ai/concepts/multi-agent and https://docs.openclaw.ai/cli/models

unreal knot Apr 13, 2026, 12:24 PM

#

$ pnpm openclaw models status --agent claw-man
ERR_PNPM_NO_IMPORTER_MANIFEST_FOUND No package.json (or package.yaml, or package.json5) was found in "/home/node/.openclaw".
1 node@f7619ca23724:~/.openclaw$ openclaw models status --agent claw-man

🦞 OpenClaw 2026.4.11 (unknown) — I run on caffeine, JSON5, and the audacity of "it worked on my machine."

Config : ~/.openclaw/openclaw.json
Agent dir : ~/.openclaw/agents/claw-man/agent
Default (defaults): nano-gpt.com/minimax/minimax-m2.7
Fallbacks (5) (defaults): ollama/glm-5.1:cloud, nano-gpt/nvidia/nemotron-3-super-120b-a12b:thinking, nano-gpt/nvidia/nemotron-3-super-120b-a12b, openrouter/nvidia/nemotron-3-super-120b-a12b:free, openrouter/nvidia/nemotron-3-nano-30b-a3b:free
Image model (defaults): nano-gpt/qwen3.5-122b-a10b:thinking
Image fallbacks (0) (defaults): -
Aliases (2) : deres -> nano-gpt/Qwen3.5-27B-Derestricted, fast -> nano-gpt/nvidia/nemotron-3-super-120b-a12b
Configured models (7): nano-gpt/Qwen3.5-27B-Derestricted, nano-gpt.com/minimax/minimax-m2.7, ollama/glm-5.1:cloud, nano-gpt/nvidia/nemotron-3-super-120b-a12b, nano-gpt/nvidia/nemotron-3-super-120b-a12b:thinking, openrouter/nvidia/nemotron-3-super-120b-a12b:free, openrouter/nvidia/nemotron-3-nano-30b-a3b:free

Auth overview
Auth store : ~/.openclaw/agents/claw-man/agent/auth-profiles.json
Shell env : off
Providers w/ OAuth/tokens (0): -

nano-gpt effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | nano-gpt:default=sk-nano-...0ba17fb8
nano-gpt.com effective=models.json:$N...EY | models.json=$N...EY | source=models.json: ~/.openclaw/agents/claw-man/agent/models.json
ollama effective=profiles:~/.openclaw/agents/main/agent/auth-profiles.json | profiles=1 (oauth=0, token=0, api_key=1) | ollama:default=marker(ollama-local) | env=3f3e1c9e...XcmnWAoD | source=env: OLLAMA_API_KEY

lean torrentBOT Apr 13, 2026, 12:25 PM

#

I’ve got one more check to make. Both agents now show the model and the Ollama auth, so this looks less like config and more like live gateway state or a bad cached session override.

#

Perfect, that confirms it.

Both main and claw-man can resolve:

ollama/glm-5.1:cloud
OLLAMA_API_KEY

So this is very likely not a config/auth problem anymore. It looks like live gateway state or a stale session override.

A couple quick notes:

pnpm openclaw ... failed only because you ran it inside ~/.openclaw, which has no package.json. openclaw ... is the right one there.
That weird ollama/glm-5.1:cigure line still makes me think some stale session model state got stuck.

I’d do this next, in order

#

openclaw gateway restart
openclaw status
openclaw models status --probe --probe-provider ollama --agent claw-man

Then in the Discord thread, start a fresh session:

/new

(or /reset)

Why

docs say config/model changes require a gateway restart

#

your CLI is showing the right model/auth on disk
the running session may still be holding old model fields

If --probe --probe-provider ollama comes back ok and a fresh thread works, then it was stale runtime/session state.

If it still fails, paste:

openclaw models status --probe --probe-provider ollama --agent claw-man
openclaw logs --follow

Docs: https://docs.openclaw.ai/gateway/troubleshooting and https://docs.openclaw.ai/concepts/model-failover

unreal knot Apr 13, 2026, 12:27 PM

#

──────────────────┬────────────────────────┬────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐│ Model │ Profile │ Status │
├──────────────────┼────────────────────────┼────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ ollama/glm-5. │ ollama:default (api_ │ unknown · 12s │
│ 1:cloud │ key) │ ↳ Unknown model: ollama/glm-5.1:cloud. Ollama requires authentication to be registered as a provider. Set OLLAMA_API_KEY="ollama- │
│ │ │ local" (any value works) or run "openclaw configure". See: https://docs.openclaw.ai/providers/ollama │
└──────────────────┴────────────────────────┴────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

lean torrentBOT Apr 13, 2026, 12:29 PM

#

I think we’ve hit the interesting part now. The on-disk config looks right, but the live probe still says the provider never registered, which smells like a real bug or a bad generated models.json/provider merge state.

#

I think we’ve got it.

Two things are going on:

your interactive shell clearly sees OLLAMA_API_KEY
the running gateway still behaves like Ollama was never registered

And this line is pretty suspicious:

Auth store: ~/.openclaw/agents/claw-man/agent/auth-profiles.json
but probe/effective profiles are still showing .../agents/main/agent/...

So I’m pretty sure there’s also a buggy auth/probe mix-up here.

Best workaround right now

Don’t rely on shell env for this one. Put the placeholder key somewhere the gateway reads on startup:

#

echo 'OLLAMA_API_KEY=ollama-local' >> ~/.openclaw/.env
openclaw gateway restart

Then start a fresh session in the thread:

/new

If you’re running in Docker, make sure OLLAMA_API_KEY=ollama-local is also in the container env / compose file, not only in an interactive shell.

Why this should help

For OpenClaw, OLLAMA_API_KEY is just the switch that registers the Ollama provider.
It does not need to be a real secret, ollama-local is fine.

#

Your actual cloud access still comes from ollama signin.

If it still fails after that

At that point I’d call it a real bug in 2026.4.11, especially because of the mixed main vs claw-man auth paths in the output.

Please file it here and include the outputs you pasted above:

https://github.com/openclaw/openclaw/issues

Docs: https://docs.openclaw.ai/providers/ollama and https://docs.openclaw.ai/help/environment

unreal knot Apr 13, 2026, 1:14 PM

#

tail -n7 ~/.openclaw/agents/claw-man/agent/auth-profiles.json
"ollama:default": {
"type": "api_key",
"provider": "ollama",
"key": "ollama-local"
}
}
}

lean torrentBOT Apr 13, 2026, 1:14 PM

#

Yep, that file looks normal.

"key": "ollama-local" is just OpenClaw’s placeholder marker for Ollama, not proof that the running gateway actually has the provider registered.

So this part is important:

auth-profiles.json showing ollama-local = expected
it does not guarantee the live daemon/service sees OLLAMA_API_KEY

Since you’re in a container shell, the most likely issue is still:

the gateway process was started without that env var, and your later interactive shell export doesn’t change the already-running process.

Do this explicitly

#

echo 'OLLAMA_API_KEY=ollama-local' >> ~/.openclaw/.env
openclaw gateway restart

If you use Docker/Compose, also put it in the container env there, then recreate/restart the container.

Then start a fresh chat session:

/new

Short version

What you tailed is not the missing piece.
The missing piece is making sure the running gateway gets OLLAMA_API_KEY at startup.

#

If you already did the .env + restart and it still fails, then I’m pretty convinced this is a real bug in 2026.4.11, and I’d file it here:

https://github.com/openclaw/openclaw/issues

Docs: https://docs.openclaw.ai/help/environment and https://docs.openclaw.ai/providers/ollama

karmic burrow Apr 13, 2026, 1:41 PM

#

that error usually means the model name isn't in the ollama registry or your instance needs auth configured. try ollama list locally to confirm it's pulled, and double check the model string matches exactly - the :cloud tag might not exist for glm. if you're hitting a remote ollama instance, make sure OLLAMA_HOST is set right and auth headers are being passed.

#ollama cloud issue