Note: The tool mentioned in this article, formerly known as Clawdbot, has been rebranded to OpenClaw. This article has been updated to reflect the new name and installation instructions. For the latest information, please visit the official website: openclaw.ai.
When AI first appeared, many of us expected it to handle the mundane tasks of our lives—ordering groceries, managing emails, or even doing laundry. This expectation hasn’t quite materialized. Instead, AI has often been associated with generating low-quality content.
Fortunately, for those seeking real benefits from AI, things are starting to improve. A new tool has emerged that fulfills the original promise of AI: managing the boring tasks of our digital lives. This includes managing emails, automatically replying to business requests, updating accounting software, and even monitoring flight prices to find good deals.
This has been promised before. However, the difference now is an open-source piece of software running on a private server. This setup enables more interesting applications, like building and publishing the Linux version of an app on command or turning raw thoughts into concrete project tickets. Perhaps the most compelling use case is its ability to manage the very machine it runs on—installing applications, improving system security, or expanding its own capabilities.
The AI future we hoped for is here, and its name is OpenClaw.
The Dawn of a True AI Assistant
OpenClaw is an open-source agent that connects to a configured LLM provider. It handles tasks and messages sent through common messaging apps like Telegram, Discord, Slack, or WhatsApp. Beyond automating tedious tasks, it’s incredibly fun to use. The online community has given it huge praise, and for many, it’s the most enjoyable server project since the days of home labs.
There’s one small hurdle: it requires some technical skill to set up. However, the process has become significantly easier recently thanks to the dedicated developers behind it.
In this article, I’m going to show you how to set up your own OpenClaw personal assistant on a personal server. We’ll also explore some personal configuration options and delve into the concept of “skills” to empower your OpenClaw to achieve almost anything you ask of it (except, perhaps, your laundry).
Getting Started: Your Server Setup
To begin, you need a place to install OpenClaw. This can be any machine you have available, such as an old laptop, a desktop PC, or a Mac Mini. While these are viable options, for the best experience, you’ll want a machine that is always online.
Although you can achieve this with your home network, the easiest and most affordable method is to use a Virtual Private Server (VPS) instance. A VPS provides a high-resource, long-term server at an affordable price, making it perfect for this use case. An instance with two vCPUs, 8 GB of RAM, and 100 GB of SSD storage is more than enough for an AI assistant, providing headroom for building projects or running a simple web app.
With your VPS instance ready, the next step is to select an operating system. The choice is yours, but for the smoothest experience and to follow this guide closely, the latest Ubuntu LTS (currently 24.04) is the best choice.
After choosing your instance size and OS, follow the setup steps. Set a secure root password and add an SSH key. Once these steps are complete, your machine will be configured, which takes about a minute. You will then be presented with the SSH command to access your new machine. Copy that command and paste it into a new terminal window to connect.
Securing Your VPS
Now that you’ve set up your VPS and connected via SSH, it’s time to implement some good security practices.
First, create a new user on the VPS called openclaw using the adduser command.
sudo adduser openclaw
This will prompt you for a password, which you should set, and some other information that you can skip.
Next, set up SSH keys for the new user. Open a new terminal window and run the ssh-copy-id command, replacing the IP address with your VPS’s IP.
ssh-copy-id openclaw@YOUR_VPS_IP
You’ll be prompted for the password of the user you just created. Once the key is copied, you can SSH as the new user without a password.
The final step is to add this user to the sudo group to perform elevated tasks. We’ll also allow the user to do this without providing a password.
Note: This is a reckless security practice. Do not use this on a machine you care deeply about. For a dedicated VPS instance, it’s a trade-off for convenience, preventing the need to send passwords over a messaging app. A better method for handling secrets will be shown later.
To allow passwordless sudo, run sudo visudo and add the following line at the end of the file:
openclaw ALL=(ALL) NOPASSWD: ALL
With the changes made, confirm that the user can perform sudo commands by SSHing back in and running a simple command.
sudo ls
If configured correctly, this should work without a password prompt. We’ve intentionally left out a few other security practices, as we’ll delegate those to OpenClaw itself.
Installing OpenClaw
With the user set up, we’re ready to install OpenClaw.
Note: OpenClaw now offers a single command to install both dependencies and the application, simplifying the process.
curl -fsSL https://openclaw.ai/install.sh | bash
After a minute or two, OpenClaw will be installed, and we’re ready to set it up.
Onboarding Your Digital Assistant
Setting up OpenClaw is relatively simple thanks to its interactive wizard. Access it by running the openclaw onboard command inside the VPS.
openclaw onboard
This will take you into the onboarding TUI (Text-based User Interface).
- Onboarding Mode: You’ll be given a choice between
Quick StartorAdvanced. For this article, we’ll chooseQuick Start. - LLM Provider: OpenClaw supports several providers, including Anthropic, OpenAI, Google Gemini, and others. I will use my ChatGPT Plus plan by selecting
OpenAI Codex. This will prompt you to load a URL in your browser to sign in. Once completed, you’ll land on a page that might suggest an error, but it’s correct. Copy the URL of this page and paste it back into the TUI to authenticate OpenClaw. This process feels a bit janky because it hijacks the OAuth flow. - Messaging Provider: You can select Telegram, WhatsApp, Discord, Slack, and more. For now, skip this; we’ll add it later.
- Skills: The next prompt asks if you want to configure skills. We’ll look at skills later, so for now, select
no.
With that, the onboarding TUI is complete. You can now communicate with your new digital assistant to finish the setup. Use the openclaw tui command to open a simple chat box.
openclaw tui
Start by saying “hello.” The bot will respond with a message asking for information to customize itself. I’ll name my assistant Zenbot and give it the persona of a Zen fox with a fox emoji. Lastly, provide your name and a nickname.
You should receive a confirmation that the assistant has removed its onboarding step and saved the new information. You might also see a message that it was unable to restart due to restarts being disabled. You can either restart it manually, enable restarts in the config, or simply ask your assistant to do it. This step can be hit-or-miss; sometimes it works perfectly, other times you may need to give it a kickstart by restarting the TUI session.
Enhancing Security with AI
With the bot configured, it’s time to address the remaining security issues on our VPS, specifically related to SSH. Rather than figuring these out by hand, let’s assign this task to our new assistant.
We’ll use the web UI for this. By default, the UI is not available over the internet, as it’s bound to localhost. To access it, set up an SSH tunnel in a new terminal window.
ssh -L 18789:localhost:18789 openclaw@YOUR_VPS_IP
Replace the IP with your VPS’s IP. With the tunnel running, you can access the web UI at http://localhost:18789. This dashboard provides the current health of your bot and configuration tabs.
Using the chat interface, send a message to your assistant to review the SSH configuration and suggest improvements. Be sure to specify that it should confirm with you before taking any action.
The assistant will propose changes. The proposed changes should align with standard security practices. Let it know it has sudo permissions and confirm the changes you want it to make. After about a minute, you’ll get a confirmation of the changes.
To verify, open a new terminal window:
- Try to SSH as the
rootuser. It should be disabled. - Try to SSH using password authentication. It should be rejected.
- Finally, ensure you can still SSH as your
openclawuser with public key authentication.
Everything should work as expected. Your bot has successfully upgraded the security of the machine it’s running on.
Connecting to Messaging Apps
Next, let’s add the ability to message our bot using a normal messaging app. The easiest to set up is Telegram. While you can ask your assistant to do this, you’d need to provide a bot token securely. For this article, we’ll use the OpenClaw CLI.
Run the openclaw provider add command on your VPS.
openclaw provider add
Select Telegram from the list, choose the default account, and you’ll be prompted for your bot token. To get a token, message the BotFather on Telegram and create a new bot. Copy the token and paste it into the terminal UI. You can accept the defaults, which require you to explicitly approve any pairing requests.
With Telegram set up, authorize your personal account by initiating a chat with your new bot. You’ll receive a message with a command to run on your VPS, including your user ID, to accept the request. Paste that command into your VPS terminal.
Now, you should be able to message the bot over Telegram and receive a response.
Unlocking Capabilities with Skills
With a simple messaging interface, it’s time to put your bot to work. A great place to start is by adding skills. These are Markdown documents that provide instructions for an agent on how to use a tool or perform a task. Think of them as single recipe cards that narrow down the choices an LLM can make.
The best place to find skills is the official OpenClaw skill website, ClaudeHub, where community members upload their creations. These skills are stored within OpenClaw’s file system, which OpenClaw can manage itself.
Let’s add our first skill using OpenClaw: the ClaudeHub skill itself. This will give our assistant the ability to search for and download skills from ClaudeHub. Ask your bot to install this skill by passing in the URL and ensuring it sets up any required dependencies.
After a minute, the bot will confirm the setup. You can verify this by running openclaw skills list on the VPS. With that, you can begin adding skills from ClaudeHub.
Automating Email Summaries
Let’s use this new capability to add more functionality. Suppose you want the bot to summarize new emails from your Gmail account. To achieve this, we’ll add the gog skill, which provides instructions on using the gog CLI to access Google services.
Ask Zenbot to search for and install the gog skill and any dependencies, this time without providing a URL.
You might find that the skill is added, but the gog binary is not, especially if the skill is optimized for macOS and relies on Homebrew. If the bot suggests an alternative installation method, like using Go, follow its lead. It might even download the binary directly from GitHub if it encounters issues.
Next, you need to authenticate with Google by setting up an OAuth client in the Google Developer Dashboard. Once you have the client details, you can securely copy them to your server and ask the bot to authenticate. You’ll provide your email address, and it will generate an OAuth URL for you to click, log in, and paste back the redirect URL, similar to the OpenAI authentication.
You may encounter an error if you haven’t set up a keyring password. This is a good security measure. You can resolve this by securely copying over a password file and trying again.
Once authenticated, test it by asking the assistant to summarize the last email you received. After a minute or two, you should get back a summary.
Advanced Automation Triggers
Now that your bot can summarize emails, you can set up a repeating interval, or a cron task, to check your email twice a day and notify you of anything important.
This highlights a key feature: OpenClaw can be triggered in several ways.
- Messages: Responding to direct messages.
- Cron: Scheduling tasks to run periodically. For example, you could have a cron task check for flight prices using an API and notify you of good deals.
- Webhooks: Exposing a small HTTP endpoint for external triggers. This allows for even more automation, such as receiving a notification whenever a purchase is made on your e-commerce store.
- Auth Changes: Monitoring for auth changes, which is useful for notifications about expiring credentials.
- Gmail Pub/Sub: Responding to email events directly.
- Polling: Periodically checking a task.
These triggers make OpenClaw extremely versatile. The only limit is your imagination and any real-world constraints.
Final Tips and Best Practices
OpenClaw is not perfect. It can sometimes feel like it’s held together with duct tape. Even so, it’s incredibly fun and adds a new dimension to automation and self-hosting.
Here are some tips:
- It’s an LLM: It will sometimes do the wrong thing. Do not run this on production hardware. Use an instance you are comfortable with breaking.
- Use Git: Use Git and the GitHub skill to back up and version control your bot’s configuration. This is invaluable for creating backups or reverting breaking changes. Just make sure to ask it not to commit any secrets.
- Use a Secrets Manager: This is the most important tip. Use a secrets manager like Doppler to share passwords with your bot. You can set up a skill that allows you to add secrets remotely without storing them in a file or sending them over Telegram. This is done by providing a limited-access token for a specific project in your Doppler configuration.
OpenClaw is a fantastic and affordable option for anyone looking to dive into personal AI automation. It’s reliable, affordable, and a great platform for running projects like this.