Running OpenClaw Locally With Ollama

A quick note before we begin.

This was supposed to be a live meetup session. I had the slides ready, the demos prepped, the whole thing. Then I started digging deeper into the security situation around OpenClaw and decided I wasn't comfortable running it on my own system, let alone asking thirty people to install it on theirs during a live demo. So I cancelled the event.

But the research was done. The notes were written. And the topic is genuinely useful if you understand the risks and make your own informed decision. So here it is, for free, as a thorough written guide. No paywall. Partake at your own risk.

What We're Building

OpenClaw can use Ollama for local inference. Messaging platforms, web search, downloads and other integrations still communicate over the network. Local model hosting is only one part of the data path.

Local inference avoids a per-request model API fee. Network access and integrations still need to be configured deliberately.

Why run the model locally?

Local inference gives you control over the model and where its requests are processed. It does not disable network access in the agent or its tools.

What This Tutorial Covers

We're going end to end. Ten files. Every step documented.

1. This overview. Why we're here, what the risks are, what you're
   getting into.

2. Hardware requirements. What your Mac actually needs to run a 14B
   model without melting.

3. Installing Ollama. The local model server that handles GPU
   acceleration on Apple Silicon.

4. Choosing and pulling a model. Not all 14B models are equal, and
   the wrong choice will waste your time.

5. Installing OpenClaw. Three methods, each with their own gotchas.

6. Configuration. The JSON config file that tells OpenClaw to use
   Ollama instead of cloud providers.

7. Security hardening. This is the big one. The ClawHavoc supply
   chain incidents demonstrated why extensions need review. We cover
   everything you need to lock down.

8. Using OpenClaw. Dashboard, terminal UI, command line. No
   messaging platforms required.

9. Troubleshooting. Every common error, why it happens, how to
   fix it.

10. Performance tuning and testing. Getting the most out of Apple
    Silicon's unified memory, and verifying the whole stack works.
+------------------------------------------------------------------+
|  USER DISCRETION ADVISED                                         |
|                                                                  |
|  This guide documents how to install third party software that   |
|  has had serious security incidents. Read the security section   |
|  before installing anything. Make your own risk assessment.      |
|  The author cancelled the live event over these concerns.        |
+------------------------------------------------------------------+

The security context

OpenClaw can act with the permissions of the account running it. Review the project security guidance and installed-version advisories before use. CVE-2026-25253 is one documented example of a gateway-token and code-execution vulnerability. Historical incident counts are not a substitute for checking the version you run.

What You Need

Hardware:

Apple Silicon Mac (M1, M2, M3, or M4)
Minimum 16GB unified memory (24GB or higher strongly recommended)
At least 15GB free disk space
macOS Sonoma (14) or newer

If you have an Intel Mac or 8GB of RAM, this won't work. Check your specs: Apple menu, About This Mac.

Software (we'll install everything in the following sections):

Ollama (local model server)
OpenClaw (the AI agent framework)
Node.js 22.12+ (OpenClaw dependency, auto-installed by the installer)

Time: About 15 minutes if you pre-download the model About 30 minutes if downloading live on a decent connection

Let's get started.