05-install-openclaw.txt
From: Running OpenClaw Locally with Ollama on Apple Silicon
INSTALLING OPENCLAW
====================
OpenClaw is the agent framework that sits on top of your model. Ollama
is the engine; OpenClaw is everything else. It handles conversations,
tool execution, memory, sessions, and optionally connects to messaging
platforms. You can think of it as the brain and personality layer.
Three installation methods, all produce the same result.
Method A: The One-Liner (Recommended)
--------------------------------------
curl -fsSL https://openclaw.ai/install.sh | bash
This is the simplest path. The script detects your system, installs
Node.js if needed (version 22.12+), and installs the latest OpenClaw
release. It handles dependencies automatically.
After it finishes, verify:
openclaw --version
Method B: npm Direct Install
------------------------------
If you already have Node.js 22.12+ installed:
npm install -g openclaw@latest
Check your Node version first:
node -v
If it's below 22.12, you need to update Node before this will work.
The one-liner method handles this for you, which is why it's the
recommended approach.
Method C: Homebrew Cask
------------------------
brew install --cask openclaw
This installs OpenClaw as a native macOS application. It requires
macOS 15 (Sequoia) or newer. If you're on Sonoma, use Method A or B
instead.
The macOS Gatekeeper Problem
------------------------------
If you use Method C (the Homebrew cask), you might hit the Gatekeeper
warning: "OpenClaw cannot be opened because Apple cannot check it for
malicious software."
This is a standard macOS security popup for apps downloaded outside
the App Store. Two fixes:
Fix 1: Right-click the app, click Open, then click Open again on
the confirmation dialog.
Fix 2: Remove the quarantine attribute from the command line:
xattr -cr /Applications/OpenClaw.app
Both do the same thing. The quarantine attribute is macOS's way of
flagging downloaded apps. Once you clear it, the app opens normally.
Running The Onboarding Wizard
-------------------------------
After installation, run the onboarding:
openclaw onboard --install-daemon
The wizard walks through several setup steps:
1. AI Provider: Select "Ollama" when it asks. Do not enter any cloud
API keys (Anthropic, OpenAI, OpenRouter). We want fully local.
2. Gateway settings: Accept the defaults or customize the port. The
gateway is the local web server that handles OpenClaw's UI and API.
3. Messaging platforms: Skip these for now. You don't need WhatsApp
or Telegram to use OpenClaw. We'll cover the built-in interfaces
that require no external services.
4. Daemon: The --install-daemon flag sets up a background service
so OpenClaw's gateway starts automatically. You can skip this and
start it manually each time if you prefer.
The wizard creates the configuration directory at ~/.openclaw/ and
writes an initial openclaw.json config file.
Alternative: Ollama Integrated Launch
---------------------------------------
If you have Ollama v0.17.0 or newer, there's a shortcut:
ollama launch openclaw
This handles installation and initial configuration in one step,
pre-configured to use your local Ollama instance. However, the manual
configuration in the next section gives you more control and is easier
to debug when things go wrong. For a meetup or learning scenario, I
recommend the manual approach.
Verify The Installation
------------------------
openclaw --version
openclaw doctor
The doctor command checks the overall health of your OpenClaw
installation: configuration validity, provider connectivity, gateway
status, and dependency versions. It's the first thing to run when
something seems off.
+----------------------------------------------------------+
| After installing, do NOT send your first message yet. |
| The default configuration may try to reach cloud APIs. |
| Configure local-only operation first (next section). |
+----------------------------------------------------------+
If the installation succeeded and doctor shows no critical errors,
move on to configuration. The next section is where we make sure
OpenClaw talks only to your local Ollama instance and never phones
home to any cloud provider.