How to Set Up and Secure OpenClaw on a VPS

TL;DR

OpenClaw can be installed on a Linux VPS, connected to an AI model and Telegram, then instructed conversationally to build useful tools such as a personalized news briefing and live server-monitoring dashboard. It is an AI harness rather than an AI model, and its broad access creates serious risks, including prompt injection, malicious skills, and unsafe server actions, so security configuration is essential.

Transcript

Okay, here we go. Open Claw. We're going to need two copies for this. This software is the reason AI has been stressing me out so much. See this video here, but you can't ignore it, even though I want to, even though I kind of hate it. It's here. Currently, it has 308,000 GitHub stars, beating React and the Linux kernel somehow. The creator got aqu... Read More

Key Insights

• OpenClaw is an AI harness, not an independent AI model. It provides a layer around an external model, letting users select a cloud model from OpenAI or Anthropic or use a locally hosted model through officially supported Ollama integration.
• The OpenClaw Gateway can run on a Linux VPS rather than dedicated hardware such as a Mac Mini. The demonstrated workflow connects to the server through SSH, retrieves the current one-line installation command from OpenClaw's documentation, and completes the remaining setup interactively.
• OpenClaw supports flexible model authentication. The setup can use a metered API key or an existing qualifying ChatGPT subscription, followed by a browser-based authorization flow and selection of the default model offered during configuration.
• Telegram acts as a remote communication channel for the agent. Creating a bot through BotFather provides the required token, while OpenClaw's synchronization step prevents arbitrary Telegram users from immediately communicating with the newly configured agent.
• OpenClaw can configure its identity and behavior through conversation. Instructions describing the agent and user are written into configuration files such as soul.md, allowing personality and operating context to persist beyond the initial terminal exchange.
• A personalized news briefing can be created from a single detailed request. The demonstrated agent checks cybersecurity content from Reddit, Hacker News, and YouTube, judges whether each item deserves attention, and turns its findings into an accessible dashboard.
• An IT monitoring dashboard can be generated by asking the agent to inspect its own server. The example checks internet speed, RAM, CPU, security conditions, and logs, then builds a dashboard that updates live without a manually constructed monitoring workflow.
• OpenClaw's broad capabilities create meaningful security risks. The source specifically warns about prompt injection, malware hidden in skills, and unsafe server access, making built-in security audits, restricted tool profiles, explicit red lines, and careful skill selection important parts of deployment.

Questions & Answers

Q: How do you install OpenClaw on a Linux VPS?

Start with a Linux VPS and copy its SSH command into a terminal to connect to the server. Visit OpenClaw's official website because its documentation can change, locate the current one-line installation command, and run it on the VPS. After installation, acknowledge the security warning, choose the quick-start process, select an AI model provider, authenticate, and continue through the interactive configuration.

Q: What is OpenClaw, and does it include its own AI model?

OpenClaw is a harness that sits on top of other AI models rather than an AI model by itself. During setup, the user chooses the agent's underlying brain. The source demonstrates OpenAI and mentions Anthropic as another option, while local models through Ollama are officially supported. This separation lets the same OpenClaw environment use different model providers and authentication methods.

Q: How do you connect OpenClaw to an OpenAI account?

Select OpenAI during the interactive quick-start process, then choose either an API key for pay-as-you-go access or the option tied to an eligible existing ChatGPT subscription. For subscription authentication, copy the displayed authorization URL into a browser, sign in, copy the resulting redirect URL from the browser address bar, and paste that URL back into the terminal.

Q: How do you connect an OpenClaw agent to Telegram?

Open Telegram and begin a conversation with BotFather, then create a new bot, assign it a name, and choose a username that ends in "bot." Copy the bot token supplied by BotFather and paste it into OpenClaw's setup prompt. After starting a chat with the new bot, complete OpenClaw's synchronization step so the authorized Telegram account can communicate with the agent remotely.

Q: What hooks should be enabled during OpenClaw setup?

The demonstrated setup enables the boot, bootstrap, command logger, and session memory hooks. The presenter skips Brave configuration after it requests an API key and postpones skill configuration. These choices produce a quick initial setup while preserving startup behavior, command logging, and session memory. They are presented as getting-started settings rather than a complete hardened production configuration.

Q: How can OpenClaw create a personalized news briefing?

Give the agent a detailed request describing your interests, preferred sources, and desired evaluation criteria. In the example, the agent is asked to find cybersecurity news, inspect Reddit, Hacker News, and YouTube, and decide whether each item is worth reading or watching. It gathers the material, rates its usefulness, and then creates a dashboard for reviewing the resulting briefing.

Q: How can OpenClaw monitor the server where it is installed?

Begin a new context and tell the agent to act as an IT engineer responsible for monitoring its own server. The demonstrated request asks it to inspect internet speed, RAM, CPU, security information, and logs, then create a dashboard. The agent first investigates the VPS to avoid interfering with existing components, builds the monitoring interface, and produces a live-updating view.

Q: What security risks should be addressed before relying on OpenClaw?

OpenClaw should be treated cautiously because an agent with server access can be exposed to prompt injection, malicious code hidden in skills, and unsafe actions affecting the VPS. The description recommends using the built-in security audit, defining restricted tool profiles, establishing red lines, and carefully configuring memory and tools. Users should not interpret a fast successful installation as proof that the deployment is secure.

Summary & Key Takeaways

• OpenClaw runs as a gateway on a server and places a conversational agent around a selected AI model. The demonstrated setup uses a Linux VPS, OpenClaw's current installation command, an OpenAI subscription, Telegram, selected hooks, and the terminal interface. Configuration continues naturally by speaking directly to the newly created agent.

• The configured agent creates a personalized cybersecurity news briefing by checking sources such as Reddit, Hacker News, and YouTube, rating whether items deserve attention, and presenting the results through a dashboard. This conversational, one-shot process replaces a more elaborate n8n workflow that previously required nodes, programming, and considerable setup time.

• OpenClaw can also inspect its own VPS, review internet speed, memory, processor use, security information, and logs, then create a live monitoring dashboard. That power makes security critical. Users should understand its models, channels, memory, and tools, run its security audit, restrict tool profiles, establish red lines, and treat third-party skills cautiously.