Setting Up A Fully Local OpenClaw

Introduction

OpenClaw is a powerful AI tool that allows your AI agent to essentially use your computer for you. However, it starts much like a newborn, where you have to teach it what to do yourself. While this may seem daunting, we have the ability to reset it as much as we want. This means trying multiple setups until one feels right for you. In this post, I want to tell you how to get OpenClaw up and running fully locally, using Ollama as a model provider. This means you are free to try things out yourself without fear of incurring costs using paid LLM models online.

[!TIP] 💡 Tip I use Ollama here as it is the easiest to setup. However there are other local AI model providers that allows better customability. When you get more comfortable, An example is llama.cpp from which Ollama is derived from. It comes with extras that allows you to modify how the model uses your computer. You should read up more about it if you are interested in optimizing your model for your computer.

What You Need

Before starting, make sure you have:

A Mac or Linux machine
Access to Terminal

[!WARNING] ⚠️ For Windows Users It’s better to run this through WSL (Windows Subsystem for Linux) rather than native PowerShell

Setting Up Ollama

Downloading Ollama

First, let’s install Ollama to your system. Open Terminal on your system and run:

curl -fsSL https://ollama.ai/install.sh | sh

and wait until it’s done. This installs Ollama and it’s requirements to your system. To confirm that Ollama was installed successfully, run:

ollama --version

in your Terminal. It should output somthing like this:

ollama version is x.xx.x

If that doesn’t work, try closing the Terminal and opening it again, as the command might have not stuck the first time round.

Downloading a model

For a local OpenClaw setup, you want something small enough to run easily but capable enough to respond reliably. A good starting option is Qwen 3.5. You can download it by using this command in your terminal:

ollama pull qwen3.5:9b

This is a lightweight model, which makes it ideal for testing the overall setup without needing powerful hardware. Once it finishes downloading, verify it’s available in Ollama by typing the command:

ollama list

You should see qwen3.5:9b in the list.

[!TIP] 💡 Tip Qwen 3.5 is a good starting point, but might not be fully suitable for your needs. Use it to get OpenClaw up and running, and try it out first. If that works for you then great! Otherwise you can try larger models. For a list of models available in Ollama, visit the Ollama Models page.

To test that Ollama is responding, try:

ollama run qwen3.5:9b

If it opens an interactive prompt, that means Ollama is working properly. You can chat with it like any other AI chatting service at this point. And once you are done playing around, type:

/bye

to end the session. Ollama is installed and ready for OpenClaw now.

Setting Up OpenClaw

Dowloading OpenClaw

First thing first, you will need to install Node.js as that is part of the requirement for OpenClaw. To do so, you can follow the instructions on the Node.js website. You can check with:

node --version

in your Terminal to see if it was installed properly. Once Node.js is installed, you can install OpenClaw by running this in your terminal:

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

This will begin downloading and installing OpenClaw and it’s other requirements on your system. Once done, we can continue to configure OpenClaw for local use.

OpenClaw Setup

Once installation is completed, OpenClaw wil automatically start it’s initial setup. If for some reason it did not run, you can paste this in your terminal to start the setup:

openclaw onboard --install-daemon

Read the Security disclaimer, and select Yes to continue.
OpenClaw will then ask you to choose a Model provider. We are using Ollama here, scroll down to it and press Enter.
In Ollama Mode, choose Local Only as we want to use the model we downloaded just now and press Enter.
Input your Ollama base URL here. If you did not change anything while installing Ollama you can press Enter.
OpenClaw will the load your Ollama model list. If you were following this guide, we would only have ollama/qwen3.5:9b right now so go ahead and select that and press Enter.
Next is the Select channel setup. Here you can choose multiple methods to send chat messages to OpenClaw. For this tutorial, we will be using Telegram, so go ahead and choose that and press Enter.
OpenClaw will then tell you how to setup a Telegram bot and get it’s token. Go ahead and follow it’s instructions, and enter the Telegram bot token to continue.
Next OpenClaw can help you setup a Search provider. This is useful if you already have an API key for the listed services, or plan to self host SearXNG to do the web search for you. For the sake of this tutorial, you can select Skip for now. Here’s the link to my SearXNG setup guide when you are ready to add local web search.
After that, OpenClaw will ask you to Configure skills. Skills are things that OpenClaw can do beyond the scope of your local computer. They are essentially tools OpenClaw uses to interact with different tools. Not that some of them will require a paid subscription. You should go through the list and see if any of the skills catches your fancy. For this tutorial sake, we will choose Skip for now. As with the Search provider there will be a separate post for skills later.
In the next few series of questions, OpenClaw will ask for multiple API keys for skills and tools it can use. As our target this time is only a fully local installation, you can just select No to all.
In the next screen, OpenClaw will ask you whether you want to install hooks. I recommend selecting all of them for optimal OpenClaw efficiency. This is however also up to you. Read what each option do, and press Enter to continue.
And now you’re done with the basic OpenClaw setup! The next part is personalisation of OpenClaw to make it trulu your own. You can select Hatch in Terminal to do it right away, or you can do it later in the chat via OpenClaw Dashboard with the keywords:

Wake up, my friend!

Personalizing OpenClaw

When you start personalising OpenClaw, it will read BOOTSTRAP.MD in your system and ask you a few questions regarding itself. Answering OpenClaw with your preferences and it will update it’s memory files to remember your name, preferences, and what you want to do with them. These are written as IDENTITY.MD which contains OpenClaw’s personality, USER.MD which is what OpenClaw will know about you, and SOUL.MD which is it’s tasks The first information OpenClaw should ask you is:

Who am I? (What the OpenClaw agent will call itself (IDENTITY.MD))
Who are you (What the OpenClaw agent will call you (USER.MD))

After which OpenClaw will ask further questions for fine tuning to how it responds about it’s characteristics (IDENTITY.MD):

What should I act as?
What kind of replies are you looking for?
What emoji do you want me to use to respond to you?

And finally what you will be asking OpenClaw to work on (SOUL.MD):

Main purpose?
Focus on speed or accuracy?
Permissions for OpenClaw whether to work on something in your system or to access the web

This completes the BOOTSTRAP.MD instructions and you would have your own personal AI assistant in OpenClaw!

[!NOTE] ℹ️ Note If you were doing this in the Terminal UI, you can exit by pressing Ctrl+C twice. This will give you further information about your setup and what you might need or what you might have missed.

So what can you do with OpenClaw? Well, at the start OpenClaw is just a normal chat agent personalised to you, and it’s setup would help connect itself with chat services you use such as Telegram and Discord.However with it’s powerful skills, OpenClaw is open-ended and infinitely customisable. For example, you can get it to search and compile data either through the web or in your desktop, or get it to send you daily news whenever you start your day!

Conclusion

With your OpenClaw setup, the AI workflow ocean is now available in your computer. It can be as simple as you want it to be, or with a powerful enough computer you can stretch it to simplify your life as much as possible. For a practical next step, try getting OpenClaw to fetch news on a schedule.

I hope that this tutorial has helped you set OpenClaw up.

Feel free to contact me if you need further consultation for all things AI usage.