Do you want a private, powerful AI assistant without paying a monthly subscription? You’re in the right place. Today, I’m going to show you how to set up OpenClaw with Ollama.
Whether you are using a high-end gaming PC, a Windows 11 laptop, a MacBook, or a Linux machine, you can run a private, “agentic” AI right now. For this guide, I’ll be using a popular Ubuntu-based system called Pop!_OS, but the steps are very similar for everyone.
The Hardware: A Quick Reality Check
Before we start, let’s talk about specs. I’m using my HP gaming laptop for this. It’s a solid machine, but it’s definitely not a supercomputer.
AI performance depends mostly on your hardware. To have a smooth experience, you should have:
- RAM: At least 16GB.
- GPU: An NVIDIA GPU with at least 4GB of VRAM.
If you meet these requirements, you can run models like Qwen 2.5 or DeepSeek R1 locally with no issues. If your laptop is a bit older, don’t worry. I have a cloud-based solution for you later in this post.
Installing the Engine (Ollama)
The first thing we need is Ollama. Think of this as the “engine” that runs the AI models on your computer.
- Download: Open your browser and go to the official Ollama website.
- Install: Copy the install command and paste it into your terminal.
- Check: If you are on a Mac, just download the DMG file and drag it into your Applications folder. Windows users can run the installer similarly.
Once it’s installed, Ollama should automatically detect your NVIDIA GPU. To make sure everything is running correctly, type the command sudo systemctl status ollama to check its status in your terminal. You can even open the local URL 127.0.0.1:11434 in your browser to confirm it’s “Ollama is running.”
Installing the Brain (OpenClaw)
Now that the engine is ready, we need the “brain.” This is OpenClaw. While a standard AI just chats, OpenClaw is an agentic tool. This means it can actually perform tasks and connect to different services.
To install it:
- Linux/macOS: Use the one-line command provided on the OpenClaw site.
- Windows: Run the specific command inside PowerShell.
When you run the script, it does the heavy lifting for you. It checks your Node.js version and sets up the system for onboarding.
Pro Tip: For security, it’s always a good idea to install these tools on a Virtual Machine (VM) if you’re just testing things out.
The “No-Hardware” Solution (Ollama Cloud)
Here’s the catch: if you don’t have 4GB of VRAM, running models locally might be too slow. But you don’t have to quit.
You can use Ollama Cloud instead. It has a free tier that lets you run powerful models on high-end hardware for $0.
- During the OpenClaw setup, select the Ollama option.
- Choose Cloud + Local mode.
- Log in with your Google or email account to connect your device.
- I recommend choosing Kimi K2.5. It’s incredibly fast and great at following complex tasks.
Then skip rest of the options like channel setup, Skills and Hooks. Once the setup finishes, look for the Control UI URL and your token. Save these! You’ll need them to access your dashboard. You can hatch open claw in Terminal Or WEB UI.
Going Fully Local with Qwen 2.5
If you have a powerful system and want 100% privacy, you should run everything locally.
First, stop your OpenClaw gateway in the terminal.
openclaw gateway stop
Then, head to the Ollama model library and search for Qwen 2.5. It is one of the most efficient models available right now. If you have extra RAM, you could even try QWEN 3 Coder / GLM 4.7.
To pull the model, use the command: ollama run qwen2.5
To launch OpenClaw with this specific local model, use: ollama launch openclaw --model qwen2.5
Now, your AI is running entirely on your own hardware. It might be a little slower than the cloud, but your data never leaves your room.
Final Thoughts
Running your own AI assistant doesn’t have to be expensive or complicated.
- Use the Cloud if you want speed and don’t have a high-end GPU.
- Go Local if you want total privacy and have a powerful Mac or NVIDIA setup.
You now have a fully functional AI assistant at zero cost!