OpenClaw: Simple VPS Setup Guide
By NeuralNine
Key Concepts
- Open Claw: An AI agent capable of controlling an entire operating system, including file management, command execution, and communication via messaging platforms.
- VPS (Virtual Private Server): A remote, isolated server environment used here to host the AI agent.
- NVM (Node Version Manager): A tool used to manage and install specific versions of Node.js (required version: 24+).
- Bot Father: The official Telegram bot used to create and manage custom bots and generate API tokens.
- SSH Tunneling: A method to securely map a remote server port (e.g., 18789) to a local machine, allowing access to the Open Claw web dashboard.
- Identity.md & Soul.md: Configuration files that define the AI agent's personality, behavioral guidelines, and operational boundaries.
1. System Prerequisites and Setup
The installation requires a clean Ubuntu server. The process follows these steps:
- System Update: Run
sudo apt updateandsudo apt upgradeto ensure the environment is current. - Node.js Installation: Use NVM to install Node.js version 24, as standard Ubuntu repositories often provide outdated versions.
- Command:
curl(from NVM GitHub) followed bynvm install 24andnvm use 24.
- Command:
- Open Claw Installation: Install the package globally using npm:
npm install -g openclaw@latest.
2. Onboarding and Configuration
The openclaw onboard --install-daemon command initiates the setup wizard:
- Model Provider: The author utilized OpenAI (GPT-5.5) as the "brain" of the agent.
- Communication Channel: Telegram was selected for its ease of use.
- Bot Creation:
- Contact "Bot Father" on Telegram.
- Use
/new botto generate a unique name and username. - Copy the provided API token into the Open Claw setup prompt.
- Search Provider: DuckDuckGo was chosen for its lack of API key requirements.
- Hatching: The "hatching" process involves defining the agent's identity. The author instructed the agent to be "professional, concise, dry, and straight to the point" to avoid unnecessary verbosity.
3. Security Warnings
The author emphasizes that Open Claw is not inherently secure. Because the agent has full system control (reading/deleting files, executing commands), it should:
- Be hosted in an isolated environment (VPS, Docker, or VM).
- Never be granted access to sensitive data like bank accounts or personal emails unless strictly necessary and audited.
- Undergo regular
openclaw security auditchecks.
4. Web Interface and Connectivity
To access the Open Claw dashboard from a local machine:
- Verify Status: Run
openclaw gateway statusto ensure the service is running on port 18789. - Retrieve Token: Read the configuration file using
cat ~/.openclaw/openclaw.jsonto find thegateway_auth_token. - SSH Tunneling: Map the remote port to the local machine:
- Command:
ssh -n -L 18789:127.0.0.1:18789 user@ipaddress
- Command:
- Pairing: After connecting via the browser, the user must run the pairing command provided in the CLI to authorize the Telegram bot for communication.
5. Real-World Application
The author demonstrated the agent's capabilities by:
- File Management: Creating a
shopping_list.mdfile and adding items via text commands. - Voice Processing: Sending a voice memo via Telegram, which the agent parsed using OpenAI’s API to update the
shopping_list.mdfile with "water and bananas." - Proactive Tasks: The agent can be configured to set reminders and perform tasks at specific intervals, effectively acting as an automated system administrator.
Synthesis
Open Claw represents a powerful, high-agency AI tool that transforms a standard VPS into an autonomous assistant. While the setup process is relatively straightforward—involving Node.js management, API integration, and SSH tunneling—the primary takeaway is the necessity of isolation. Because the agent possesses the capability to modify system files and execute arbitrary commands, users must prioritize security by running it in a sandboxed environment and maintaining strict control over the permissions granted to the agent.
Chat with this Video
AI-PoweredLoad the transcript when you're ready to chat so the initial page stays lighter.