New: Skill bounties are live. Post a request, fund the bounty, and creators compete for 7 days to build it -> See open bounties

    Guides
    openclaw
    mac-mini
    self-hosted

    Running OpenClaw on a Mac Mini: Setup and Hardware Guide (2026)

    A Mac Mini is the cheapest way to run an always-on OpenClaw agent. Here is the hardware you need, how to keep it running headless, and what it costs.

    July 31, 20266 min read
    Share:

    A Mac Mini has become the default box for an always-on self-hosted agent. Apple Silicon's unified memory means a base model can run local coding models that would need a dedicated GPU on a PC, and it draws a few watts at idle.

    Quick Answer: An M4 Mac Mini with 16GB unified memory runs OpenClaw comfortably with API-based models and handles 7B-14B local models via Ollama. Go to 24GB or 32GB if you want to run larger local models. Install OpenClaw normally, disable sleep, enable Remote Login for SSH access, and set it to launch at login.

    Choosing the memory configuration

    Unified memory is the constraint that matters. On Apple Silicon it's shared between CPU and GPU, so it's what determines which local models you can run.

    ConfigAPI modelsLocal modelsSuits
    16GBComfortable7B–14BMost people, API-first with occasional local
    24GBComfortableUp to ~22BLocal-first with headroom
    32GB+Comfortable30B+Fully local, larger models, parallel agents

    Storage matters less than people expect. OpenClaw itself is small. Local models are the space consumer: roughly 4GB for a 7B model, 8GB for 13B, 20GB for 30B. A 256GB Mini is tight if you're collecting models; 512GB is comfortable.

    Why memory over CPU: an M4 handles the compute fine at every tier. Running out of memory is what stops a model loading.

    Recommended skills

    T212 Dividend Tracker

    T212 Dividend Tracker

    by MasterKilo

    Analyze Trading212 dividend history, calculate portfolio yield, and forecast upc…

    $9    claude-code-audit

    claude-code-audit

    by SkillBill

    Audits your Claude Code project setup and flags what's broken, misconfigured, or…

    $7

    openclaw-web-scraper

    by BCStudio · 20

    A high-performance scraping engine with Playwright stealth, proxy rotation, and…

    Free
    Browse all Openclaw skills

    Installing OpenClaw

    Standard install works unchanged:

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

    Then configure your model provider:

    openclaw config set provider anthropic
openclaw config set api-key sk-ant-your-key

    Full detail in the OpenClaw installation guide.

    Stopping it from sleeping

    An always-on box that sleeps isn't always on. This is the step people miss.

    sudo pmset -a sleep 0
sudo pmset -a disksleep 0
sudo pmset -a autorestart 1

    autorestart 1 brings the machine back after a power cut, which matters for a headless box you're not standing next to.

    Verify with pmset -g.

    What this fixes: the agent that stops responding overnight and works again when you touch the keyboard.

    Running it headless

    Enable SSH in System Settings under General, then Sharing, then Remote Login. Or from the terminal:

    sudo systemsetup -setremotelogin on

    Then run OpenClaw inside tmux so the session survives disconnection:

    brew install tmux
tmux new -s openclaw
openclaw

    Detach with Ctrl+B then D. Reattach later with tmux attach -t openclaw.

    Why tmux: an SSH session that drops takes your agent with it. A tmux session doesn't.

    Adding local models

    Install Ollama and pull a coding model:

    brew install ollama
ollama serve &
ollama pull qwen2.5-coder:14b

    Point OpenClaw at it:

    openclaw config set provider ollama
openclaw config set model qwen2.5-coder:14b
openclaw config set endpoint http://localhost:11434

    On a 16GB Mini, a 14B model leaves enough headroom for the OS and OpenClaw. Larger models will swap and slow to a crawl.

    Full setup in OpenClaw + Ollama.

    What this gets you: zero API cost and no code leaving the machine.

    Reaching it from your phone

    An always-on agent is most useful when you can send it work from anywhere. Connect the Telegram plugin:

    openclaw plugin install telegram

    Create a bot via @BotFather, paste the token, and pair. Now you can send tasks from your phone and the Mini processes them against your local files.

    Setup detail in OpenClaw Telegram Setup.

    Installing skills

    Skills are what make an always-on agent useful rather than just available. Drop them in the skills directory:

    ~/.openclaw/skills/
├── code-reviewer/
│   └── SKILL.md
├── git-commit-writer/
│   └── SKILL.md

    code-reviewer, 764 installs, free. git-commit-writer, 232 installs, free. env-doctor, 86 installs, free.

    Browse all OpenClaw-compatible skills on Agensi.

    The full setup

    1. Install OpenClaw with the setup script
    2. Configure your model provider, API or Ollama
    3. Disable sleep with pmset
    4. Enable Remote Login for SSH
    5. Install tmux and run OpenClaw inside a named session
    6. Install the Telegram plugin for remote access
    7. Drop skills into ~/.openclaw/skills/

    For hardware requirements across all platforms, read OpenClaw System Requirements.

    Keep reading

    Frequently Asked Questions