openclaw onboard not working

[ember carbon]

I used openclaw onboard to choose a new ai model, hugging face's minimax 2.5 and it showed TypeError: Cannot read properties of undefined (reading 'trim')

[tidal hamlet]

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

[ember carbon]

i use windows

[burnt vigil]

That looks like a real onboarding bug, not you.

I dug through the code, and the crash is coming from a path where onboarding ends up calling .trim() on an undefined model ref. So the wizard is likely losing the selected model somewhere.

One important detail: MiniMax M2.5 is currently documented under the synthetic provider, not the Hugging Face provider:

• Synthetic docs: https://docs.openclaw.ai/providers/synthetic
• Hugging Face docs: https://docs.openclaw.ai/providers/huggingface

If your goal is specifically MiniMax M2.5, the working path should be:

openclaw onboard --auth-choice synthetic-api-key

That should set:

synthetic/hf:MiniMaxAI/MiniMax-M2.5

If you meant Hugging Face proper, its model refs are in the form huggingface/<org>/<model>, and the docs currently show things like huggingface/moonshotai/Kimi-K2.5, not MiniMax M2.5.

Can you paste these two things so we can pin down whether this is version-specific?

openclaw --version

and, if you have it, the full stack trace after the reading 'trim' error.

If you want, I can help narrow it down to the exact broken onboarding step once you paste that.

Got it. Windows itself probably is not the cause of this specific .trim() crash, it still looks like an onboarding bug.

Two quick checks would help a lot:

openclaw --version

and tell me whether you're on:

• native Windows
• WSL2

If you want the fastest workaround for MiniMax M2.5, use the Synthetic provider path instead:

openclaw onboard --auth-choice synthetic-api-key

That model is documented as:

synthetic/hf:MiniMaxAI/MiniMax-M2.5

Windows docs are here too: https://docs.openclaw.ai/platforms/windows
The FAQ currently recommends WSL2 over native Windows for the smoothest setup.

If you paste your version next, I can tell you whether this looks like a known bug in that build.

[ember carbon]

Tried the kimi2.5 still shows the same undefined

[burnt vigil]

Yep, that pretty much confirms it, this is not model-specific. It looks like the interactive Hugging Face model picker in openclaw onboard is the broken part.

Fast workaround: set the model in config manually instead of picking it in the wizard.

If you want HF Kimi 2.5:

{
  agents: {
    defaults: {
      model: { primary: "huggingface/moonshotai/Kimi-K2.5" },
    },
  },
}

If you want Synthetic MiniMax M2.5:

{
  agents: {
    defaults: {
      model: { primary: "synthetic/hf:MiniMaxAI/MiniMax-M2.5" },
    },
  },
}

And if you already use a model allowlist, also add the same ref under agents.defaults.models.

Docs:

• Hugging Face: https://docs.openclaw.ai/providers/huggingface
• Synthetic: https://docs.openclaw.ai/providers/synthetic

Can you still paste:

openclaw --version

If this is on current OpenClaw, it’s worth filing a GitHub issue with the stack trace because this really looks like a wizard bug.

[ember carbon]

Yeah it’s the latest

[burnt vigil]

Then yeah, I’d treat this as a real bug in the onboarding model picker.

Please file it here: https://github.com/openclaw/openclaw/issues

Include these details:

• exact output of openclaw --version (latest is a bit too fuzzy)
• Windows native or WSL2
• the selected model (huggingface/moonshotai/Kimi-K2.5 or whatever you picked)
• the full stack trace with reading 'trim'
• that it reproduces with multiple Hugging Face models, not just one

For now, the workaround is still to set the model manually in config instead of choosing it in openclaw onboard.

If you paste the exact version + stack trace here first, I can help you make the bug report tighter.

[ember carbon]

That’s not only about hugging face I tried google also the same thing appeared, I think all providers are having the same problem after the update

[burnt vigil]

Ah, that’s actually really helpful, thanks.

If Google + Hugging Face + others all fail the same way after the update, then this is almost certainly a shared onboarding/model-selection regression, not a provider-specific bug.

I’d file it as:

openclaw onboard crashes with TypeError: Cannot read properties of undefined (reading 'trim') when selecting a model after update (Windows)

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

Please include:

• exact output of openclaw --version
• Windows native or WSL2
• that it reproduces across multiple providers (Google, Hugging Face, etc.)
• exact repro steps
• full stack trace