↓ Skip to main content

You don't need a Mac Mini to run OpenClaw

·1018 words·5 mins
Gabriel Guzmán
Author
Gabriel Guzmán
A regular human. (he/him)
You don't need a Mac Mini to run OpenClaw - This article is part of a series.
Part 1: This Article

You don’t need a Mac Mini to run OpenClaw1. OpenClaw is a NodeJS app that shouldn’t use up too many system resources. It should run pretty well on limited hardware like the Raspberry Pi 3b+ that’s sitting in my closet collecting dust, but let’s try it out and see.

The OpenClaw documentation has this table:

Pi Model RAM Works? Notes
Pi 5 4GB/8GB ✅ Best Fastest, recommended
Pi 4 4GB/8GB ✅ Good Sweet spot for most users
Pi 4 2GB ✅ OK Works, add swap
Pi 4 1GB ⚠️ Tight Possible with swap, minimal config
Pi 3B+ 1GB ⚠️ Slow Works but sluggish
Pi Zero 2 W 512MB ❌ No Not recommended

So it looks like a Pi 3B+ will work, but maybe the experience won’t be great2.

OpenClaw allows you to interface with multiple different LLMs, maintains a history of your interactions, updates it’s own internal memory, and provides tools that the LLMs can call to perform work for you. It provides a chat interface where you can tell it to do things for you, and allows you to easily integrate with your pre-existing chat apps to make talking to it even easier.

Since it outsources the heavy processing to an LLM, generally hosted by a 3rd party it shouldn’t need too many resources to do it’s job.

A Mac Mini could be a good choice if you want to run OpenClaw as well as a local LLM on the same hardware, but even then I’d suggest using the Mac Mini for only the LLM and run OpenClaw on separate, less powerful hardware, so they don’t interfere with each other.

I think the main things people are looking for when they go the Mac Mini route are:

  1. Isolation - I don’t want OpenClaw running amok on my main computer, instead I want it to be isolated in a system where it can’t accidentally delete my files.

  2. Availability - I want a system that can be up and running 24/7, something that won’t shutdown when I close my laptop for the night.

  3. Control - I don’t want to be locked into the ecosystem of a single provider, I want to be able to run my own LLM on my own terms.

For 2.5 of these, a Raspberry Pi is just as good as a Mac Mini. The only thing the Pi won’t do very well is run an LLM, but I don’t think I really want OpenClaw and my LLM running on the same hardware in any case.

Let’s setup our Pi.

We’ll be using a Raspberry Pi 3b+ which has 1GB of RAM and a quad core 1.4Ghz 64bit ARMv8 CPU. This should be enough to handle OpenClaw, and satisfy requirements 1 and 2 above.

Installation
#

I installed the default Raspberry Pi OS using the raspberry pi system imager.

In retrospect, I should have chosen the “Lite” version of the OS as it comes w/out the GUI which uses a few hundred MB of RAM.

Once the OS was up and running, I basically did the same thing I describe in my other post. I’ll recap here.

  • Download + Install the latest version of NodeJS
  • Download + Install OpenClaw
  • Run the OpenClaw setup wizard

And then I was off to the races, or so I thought.

Though the initial installation went well (if slowly) I soon ran up against my first out of memory error while going through the onboarding process:

FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory

At this point, I tried a few ways to free up additional RAM. I changed the boot parameters to boot w/out the GUI, upgraded the swap to 2GB, and shut down a few unneeded services like bluetooth.

This helped a little and allowed me to complete the onboarding, but then I got stuck with the same error message while attempting to configure the Discord integration.

At this point I switched to the Lite version of the OS, reapplied my swap changes, and followed the instructions on the OpenClaw page for Raspberry Pi.

I gave up trying to configure Discord during the onboarding process, as I kept running out of RAM and having to start the whole install process over again.

I skipped that step, and was able to complete the setup and start talking to OpenClaw on the Pi 3b+.

I configured OpenClaw to use Kimi K2 from moonshot.ai as it’s main LLM. That seems to be working pretty well. I’m able to do basic things with OpenClaw, like configure crons, check webpages, get the weather, chat, open files, etc.

Anything more complex and I start to hit memory issues, even just trying to upgrade OpenClaw will crash unless I explicitly set NODE_OPTIONS="--max-old-space-size=1536" before running the upgrade.

OpenClaw Setup
#

Here’s what I configured my OpenClaw instance to do:

  1. Check my blog for new articles, and let me know when they’re available.

  2. Check-in with itself daily to see how it’s doing, CPU, Storage, RAM, etc.

  3. Check the weather in the morning and let me know if there’s anything I should know about.

Thoughts
#

While OpenClaw does run on a Raspberry Pi 3b+, it’s full of frustrations. I don’t recommend running it on this particular Raspberry Pi. 1GB of RAM is just not enough.

Future work
#

I have a few other Raspberry Pi devices collecting dust in the closet, so I think I’ll try those next. I have a Mycroft Mark II, which has a Raspberry Pi 4 with 2GB of RAM in it, and a Raspberry Pi 400 which I think has 4GB of RAM. I’ll give them both a try and see how they work out.

On this pi3, maybe I’ll try one of the other OpenClaw like options that are made specifically for less capable hardware.

These are the ones that look promising:

PicoClaw - A Go alternatvie to OpenClaw NanoBot - A small Python bot inspired by OpenClaw NullClaw - The smallest fully autonomous AI assistant, written in Zig

  1. But you probably should use something with at least 2GB of RAM. ↩︎

  2. Confirmed, the experience is not great. ↩︎

Reply by Email
You don't need a Mac Mini to run OpenClaw - This article is part of a series.
Part 1: This Article