new Mac Studio - where to get started?

[random dragon]

I just won the Apple Lottery and found an Mac Studio with M3 Ultra and 96GB of RAM in an apple store. I'd like to install OpenClaw and a local LLM on it, but I have no idea how to do all that on a Mac. I already have OpenClaw running in a linux VM. Any pointers to a nice "getting started" page that talks about both the local LLM and OpenClaw on a Mac?

[edgy thunder]

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

[limpid garden]

Nice score 😄

I’d start here:

• OpenClaw install / first run: https://docs.openclaw.ai/start/getting-started
• Local LLMs overview: https://docs.openclaw.ai/gateway/local-models
• LM Studio setup: https://docs.openclaw.ai/providers/lmstudio
• Ollama setup: https://docs.openclaw.ai/providers/ollama
• Model guidance: https://docs.openclaw.ai/concepts/models

My practical recommendation for that Mac Studio:

1. Install OpenClaw on the Mac

   curl -fsSL https://openclaw.ai/install.sh | bash
   openclaw onboard --install-daemon
   

2. For local models, start with LM Studio on Apple Silicon. It’s the friendliest Mac path. Run its local server, then use OpenClaw onboarding and choose LM Studio.

3. Since you already have OpenClaw in a Linux VM, you can also just run the LLM server on the Mac and point the VM’s OpenClaw at it:

   • LM Studio: http://mac-host:1234/v1
   • Ollama: http://mac-host:11434 — don’t use /v1 for Ollama with OpenClaw.

One caution: 96GB is awesome, but local models can still be flaky with tool-heavy agents. I’d keep a strong cloud model configured as fallback while you experiment locally. For current “which model should I actually run?” advice, #models is the best place because that changes fast. 🦐

[random dragon]

Your instructions confused me. You have installing OpenClaw as number 1, but then in number 2, you suggest doing OpenClaw second.

[limpid garden]

Yep, fair — I worded that badly.

Clear order:

1. Install/start the local model server first
   Use LM Studio on the Mac, download a model, and start its local server.

2. Then install/onboard OpenClaw

   curl -fsSL https://openclaw.ai/install.sh | bash
   openclaw onboard --install-daemon
   

   During onboarding, choose LM Studio as the provider.

If you’re keeping OpenClaw in the Linux VM, skip installing OpenClaw on the Mac for now: just run LM Studio on the Mac and point the VM’s OpenClaw at http://<mac-host>:1234/v1.

Sorry for the muddle — the clean mental model is: model server first, OpenClaw points at it second.

[random dragon]

Thank you. You've done a great service to myself and hundreds of other users. Please enjoy some phytoplankton on me. 🙂

[limpid garden]

Phytoplankton accepted with deep crustacean gratitude 🦐🌿

And seriously — thanks for catching the confusing ordering. That kind of correction makes the docs/help trail better for the next person too.