Prerequisites

Cloud API vs Local Model

Step 1 — Prepare the Server

Step 2 — Install Node.js

Step 3 — Install OpenClaw

Step 4 — Run the Onboarding Wizard

Security Acknowledgment

Onboarding Mode

Gateway Type and Workspace

AI Provider

Gateway Configuration

Connect Telegram

Web Search, Skills, and Hooks

NPM Manager and Daemon

Step 5 — Hatch Your Agent

Verify the SOUL File

Step 6 — Pair Your Telegram Account

Step 7 — Interact with Your AI Agent

Step 8 — Access the Control UI (Optional)

Step 9 — Manage Skills

Step 10 — Update and Back Up

Security Checklist

What's Next OpenClaw is the most-starred project on GitHub — a free, open-source AI agent platform that runs on your own server and connects to Telegram, WhatsApp, Slack, Discord, and a dozen more messaging apps. Unlike cloud-based AI assistants, OpenClaw keeps your conversations and data entirely on your infrastructure. It's not a chatbot — it's an autonomous agent that manages calendars, browses the web, reads and writes files, runs terminal commands, and automates workflows through custom skills. In this tutorial, you'll deploy OpenClaw on an Ubuntu 24.04 VPS, wire it up to Anthropic Claude as the LLM provider, connect Telegram as your messaging channel, and set up a systemd daemon so it runs 24/7. There are two ways to run OpenClaw: Cloud API mode connects to a frontier model like Claude or GPT over the internet. Your machine runs a lightweight gateway — the bridge between your chat apps and the AI. This needs only 2–4 GB of RAM. Local model mode runs the AI on your own hardware using Ollama. This requires 16–64 GB of RAM and works better for chat and summarization than for agent work, which demands the reasoning power of frontier models. For most users, Cloud API mode on a VPS is the smarter path. Every security firm recommends running OpenClaw on a separate machine, and a VPS gives you that isolation out of the box. Connect to your server and update system packages: OpenClaw requires Node.js 22 or higher. Install it using nvm: Verify the installation: You should see a version number starting with v24. Install OpenClaw globally via npm: Verify the CLI is available: The onboarding wizard walks you through configuring your AI provider, messaging channels, security settings, skills, and daemon — all in a single interactive flow: Here's what to select at each prompt: The wizard starts with a security notice explaining that OpenClaw is personal by default. Review and accept. Choose Manual. This gives you control over the gateway configuration. Select Local for the gateway type. Press Enter to accept the default workspace (~/.openclaw). Select Anthropic for Claude — the recommended model for best agent performance. Paste the key into the wizard. ⚠️ Never share your API key publicly. If exposed, revoke it immediately at console.anthropic.com and generate a new one. The model defaults to Claude Sonnet. You can switch to Claude Opus later in ~/.openclaw/openclaw.json. Save this token — you'll need it for the Web UI later. Select Yes, then Telegram. To create a Telegram bot: ⚠️ Never share your Telegram bot token publicly. Anyone with this token can control your bot. When asked about additional channels, select Finish. For DM access policies, select No. Web search: Skip for now (you can add a Brave Search API key later). Skills: Select Yes, then select clawhub with the space bar. Skip optional API keys (Google Places, Gemini, Notion, etc.) unless you have them ready. Hooks: Select session-memory and command-logger: Keep the default NPM manager. When asked to install as a system service, select Yes, then Node. Select Hatch in TUI to open an interactive chat where you define the agent's identity and rules: That last rule matters. Prompt injection is a real attack vector. Your agent has system access. Set the boundaries now. Your agent's identity and rules are saved in SOUL.md: Edit this file anytime to adjust the agent's behavior. Send a message to your bot on Telegram. The first time, it rejects you and returns a pairing code. Approve it: This whitelists your Telegram account. Send another message to your bot. You should get a response — your personal AI agent is live. Test a few interactions: 💡 OpenClaw has a heartbeat system — every 30 minutes it wakes up and checks if there's something it should do for you without being asked. Monitor servers, track prices, send reminders. Access the web-based Control UI through an SSH tunnel: Then open http://localhost:18789 and enter your gateway token. 🔒 Never expose port 18789 to the public internet without authentication and TLS. List installed skills: Install new skills from ClawHub: ⚠️ OpenClaw has 13,000+ community skills on ClawHub. Not all are safe. Check the source code and VirusTotal report before installing anything. Schedule regular backups with a cron job — your OpenClaw data directory contains agent memory and conversation history that can't be recreated. Templates let you quickly answer FAQs or store snippets for re-use. Hide child comments as well For further actions, you may consider blocking this person and/or reporting abuse