How to Set Up OpenClaw AI Agent and Use It Safely

Looking for a step-by-step guide on how to install OpenClaw on Linux, macOS, or Windows? Here are our answers for all of your questions!

What is OpenClaw AI Agent

OpenClaw AI is an open-source personal AI agent platform. Users describe a task, and OpenClaw invents the capabilities it needs to use like opening files, cleaning your email, using a browser, connecting to a third-party service via an API, or even writing code from scratch. If you want to know more about the story of OpenClaw AI and an alternative for it, read our article.

How Does OpenClaw AI Agent Work

You can talk to the OpenClaw AI agent via familiar apps like Telegram, WhatsApp, iMessage, Slack, or Signal. It can also connect to different AI models including Claude, ChatGPT, Kimi, and others. Of course, the more powerful the neural network, the better the performance.

OpenClaw AI agents also have memory, but it works more like preferences than a full diary. They can learn how you like to communicate or work, yet they don’t automatically store every past solution. If the AI agent figures out a useful method, you usually need to tell it to remember that approach, then next time it can repeat the result much faster and with less guidance.

What Services and Tools Exist for OpenClaw?

In just a few weeks, an entire ecosystem has started to grow around the project:

• ClawHub. A marketplace for AI skills, with 700+ available. Think of it as an npm-style hub, but for AI agent capabilities.
• Clawezy. A managed hosting service for OpenClaw if you don’t want to deal with installation and configuration. Pricing starts at about $49 per month.
• GoClaw. A mobile app that lets you manage and control your AI agent directly from your phone.
  MoltBook. A social network for bots where AI agents can interact with each other. It may sound bizarre, but it actually exists.

The speed at which these tools and services have appeared is impressive. At the same time, it raises concerns. When an ecosystem grows this quickly, quality and security often struggle to keep up.

How to connect OpenClaw to Telegram, WhatsApp & iMessage?

Connecting OpenClaw to messaging apps like Telegram, WhatsApp, and iMessage lets you interact with your AI agent in the same places you already chat every day. Instead of logging into a dashboard, you can send commands, ask questions, and automate tasks directly from your favorite messenger.

How to connect OpenClaw to Telegram

To connect OpenClaw to Telegram, first open Telegram and search for the official bot @BotFather (the one with the blue checkmark), then start a chat with it. After that, the setup only takes a few steps:

1. Create a new bot by sending the command /newbot. BotFather will ask you to choose a display name (for example, My AI Assistant), and then a username that must end in bot, such as my_ai_helper_bot.
2. Copy the token. Once the bot is created, BotFather will send you a message containing the API token and save it somewhere secure.
3. Add the token to OpenClaw. Go to your OpenClaw dashboard, paste the token into the required field, and click Create Agent.

After that, your agent will be able to operate through Telegram.

Note: There are several connection modes for a bot in Telegram. The safest option is private mode, where the bot responds only to its owner. To enable this, you need to provide your Telegram ID, which you can obtain by messaging a bot that shows your account information.

If no Telegram ID is specified, other options apply. In open mode, the bot will respond to any user without restrictions. If neither open mode nor a specific owner is set, automatic pairing is used: the first person who messages the bot automatically becomes its owner.

How to connect OpenClaw to WhatsApp

OpenClaw connects to WhatsApp Web using the Baileys library. Follow these steps to set it up:

1. Add WhatsApp to the config. Edit ~/.openclaw/openclaw.json and add the WhatsApp channel. Make sure the phone number is in E.164 international format (for example, +1 for U.S. numbers).
2. Log in via QR code. Run: openclaw channels login --channel whatsapp. A QR code will appear in the terminal.
3. Link your phone. Open WhatsApp on your phone → go to Settings → Linked Devices → tap Link a Device → scan the QR code shown in your terminal.
4. Verify the connection. Run: openclaw channels status --probe. WhatsApp should appear with linked: true and running.

Note: It’s best to use a separate phone number for OpenClaw so your personal chats stay separate from the bot’s activity.

How to connect OpenClaw to iMessage

OpenClaw also can connect to iMessage on macOS using the imsg library. To set it up:

1. Enable iMessage in the config. Turn on the iMessage channel in your OpenClaw configuration file. The agent will automatically detect and connect to the Messages app.
2. Configure group behavior. You can control which group chats OpenClaw responds to using the groups option. For example, list only the groups that should receive replies. To allow responses in all groups, include "*" in the array.

Note: Your Mac should have Messages signed in and active. OpenClaw and imsg also need Full Disk Access (to read the Messages database) and Automation permission to send messages.

How to Use OpenClaw AI Safely

OpenClaw AI assistant is not secure by default. Its security depends on how it is configured, sandboxed, and managed by the developer. Let's look at how to secure OpenClaw AI on our own.

The Main Risks of Using OpenClaw AI agent tool

Because the AI agent can act on its own, a mistake in its reasoning – or manipulated input – could lead to unintended actions, such as sending incorrect messages, launching unnecessary processes, or even deleting files.

Plugins and integrations also require caution, since third-party modules may contain unsafe logic or hidden malicious code. If the AI agent platform processes external inputs like messages or API requests, attackers might try to influence its behavior through carefully crafted prompts.

It’s easy to use OpenClaw AI via a messaging app, but it also creates more ways for things to go wrong. Security researcher Jamison O'Reilly showed how serious the risks can be: he was able to access sensitive data such as Anthropic API keys, Telegram bot tokens, Slack accounts, long message histories, and even send messages as the user. In some cases, he could also run commands with administrator rights.

Besides, OpenClaw AI has ClawdHub, it's a public skills registry. “Skills” in this context means ready-made scenarios that extend the agent's capabilities. ClawdHub also has no moderation whatsoever, and instead of skills, it was loaded with a multitude of malicious codes.

If you have any doubts about OpenClaw, you can try Sigma Browser. Its AI agent has a closed code and strict monitoring of user privacy.

What User Can Do to Using OpenClaw AI Safely

If you give OpenClaw AI assistant access to systems, data, or integrations, be sure to address your security concerns beforehand. To prevent leaks, you must adhere to certain rules:

• Run the agent in an isolated environment, such as a container or a separate virtual machine. Avoid testing it directly on your main workstation or production server.
• Grant the agent only the minimum access it truly needs to specific files, folders, APIs, or services. Avoid administrator rights, global system access, or production credentials.
• Review every plugin or skill before installing it. Read the code if possible, understand what it does, and monitor its behavior. If you wouldn’t deploy it manually, don’t hand it to an autonomous agent.
• Keep secrets separate from the agent. Use secure secret managers or restricted environment variables, and never hardcode keys, tokens, or passwords into prompts, configs, or plugins. Rotate credentials regularly.
• Add human approval checkpoints for high-risk actions. Anything affecting money, production systems, or sensitive data should require confirmation. Autonomy should be introduced gradually, not all at once.

How to Set Up OpenClaw AI Agent on PC

Setting up an AI agent on your PC might sound technical, but the process is actually straightforward once you know the steps. In this OpenClaw installation guide, we’ll walk through what you need, how to install it, and how to get your first agent running.

Preparing for Install OpenClaw AI

Before installing OpenClaw, make sure you have:

• Operating system: macOS, Linux, or Windows (WSL)
• Node.js: v18 or higher
• API key for the AI ​​model: Claude or GPT key

1. Clone the project

git clone https://github.com/openclaw/openclaw.gitcd openclaw

2. Installing dependencies

npm install

3. Setting up environment variables

Create a .env file and add your API key there:

# Using the APIYI access point
OPENAI_API_KEY=your_key_here
OPENAI_BASE_URL=https://vip.apiyi.com/v1
# Or using ClaudeANTHROPIC_API_KEY=your_key_claude_here

4. Launch OpenClaw AI

npm run start

Basic settings of OpenClaw AI

The initial setup of OpenClaw includes a handful of basic configuration choices that affect security, access, and integrations. 

Security tip: For the first launch, use sandbox mode. Once you’re comfortable, you can gradually allow full system access.

How to Uninstall OpenClaw AI

To uninstall OpenClaw, you can use the built-in openclaw uninstall utility for interactive removal (with confirmations). For a complete, silent removal, use openclaw uninstall --all --yes. If the CLI is not installed, you can manually remove the gateway service.

How to Uninstall OpenClaw from macOS

Let’s look at how to delete gateway service from your macOS. You can remove it manually using several commands. The default label is bot.molt.gateway (or bot.molt.<profile>).

Base commands:
launchctl bootout gui/$UID/bot.molt.gateway ,
rm -f ~/Library/LaunchAgents/bot.molt.gateway.plist

1. If a profile was used, replace the label and plist name with bot.molt.<profile>.
2. Remove any outdated plist files com.openclaw.* if they are present (related to OpenClaw).

Note: If profiles were used (--profile / OPENCLAW_PROFILE), you need to repeat step 3 for each state directory (default: ~/.openclaw-<profile>). In remote mode, the state directory is located on the gateway host, so steps 1-4 must also be executed there.

How to Uninstall OpenClaw from Linux

In this part of our guide you'll find how to manually disable and delete the gateway user service on Linux systems. Manual removal of the gateway service (default unit name – openclaw-gateway.service).

Base commands:

systemctl --user disable --now openclaw-gateway.service ,
rm -f ~/.config/systemd/user/openclaw-gateway.service ,
systemctl - -user daemon-reload

Note: If you installed OpenClaw in a Docker-based sandbox, you also need to remove any Docker images and containers associated with OpenClaw.

How to Uninstall OpenClaw from Windows

Here we'll tell you  how to remove OpenClaw from Windows. Manual removal of the scheduled task (default task name – OpenClaw Gateway):

Commands:
schtasks /Delete /F /TN "OpenClaw Gateway" ,
Remove-Item -Force "$env:USERPROFILE\.openclaw\gateway.cmd"

If a profile was used, delete the corresponding task name and file ~\.openclaw-<profile>\gateway.cmd.

Note: The openclaw uninstall command removes the gateway service and local data, but keeps the CLI utility itself. To remove it as well, run the following after uninstalling:  npm rm -g openclaw (for OpenClaw).