Before You Start: What You Need

OpenClaw runs as a Node.js service on your own computer. Before diving into setup, make sure you have the basics ready. You do not need a powerful gaming PC — a regular laptop or a Mac Mini will do. Many users are running OpenClaw on Raspberry Pi devices and old office machines.

A computer running macOS, Windows, or Linux (always-on is ideal, but not required for testing).
Node.js version 22 or higher installed.
A Telegram account (the easiest messaging platform to start with).
An API key from an LLM provider: OpenAI (GPT-5), Anthropic (Claude), Google (Gemini), or a local model via Ollama.
Basic comfort with the command line — you will need to run a few terminal commands.

Step 1: Install OpenClaw

OpenClaw installation is straightforward. Open your terminal and run the following commands. The installer will create a configuration directory and download the latest stable release. On Windows, use PowerShell or Git Bash.

Run: npx openclaw@latest init — this initialises a new OpenClaw instance in your current directory. The CLI wizard will guide you through basic configuration including your preferred language model and messaging platform. Choose Telegram as your first platform — it has the best integration and is the easiest to debug.

Step 2: Create a Telegram Bot

OpenClaw connects to Telegram through the Bot API. Creating a bot takes about two minutes:

Open Telegram and search for @BotFather.
Send the command /newbot and follow the prompts to name your bot.
BotFather will give you an API token — copy this. It looks like: 123456789:ABCdefGHIjklMNOpqrSTUvwxYZ.
Paste the token into your OpenClaw configuration when prompted, or add it to your .env file as TELEGRAM_BOT_TOKEN.
Send a message to your new bot in Telegram to activate it.

Step 3: Connect Your AI Model

OpenClaw needs a language model to understand your messages and decide what actions to take. You have several options, each with different trade-offs:

OpenAI GPT-5.4 — Best overall intelligence, cloud-based, costs approximately US$0.01-0.03 per interaction. Set OPENAI_API_KEY in your .env file.
Anthropic Claude Sonnet 4.6 — Excellent for detailed analysis and writing tasks, similar pricing to OpenAI. Set ANTHROPIC_API_KEY in your .env file.
Google Gemini 2.5 Flash — Good balance of speed and capability, generous free tier. Set GOOGLE_API_KEY in your .env file.
Ollama (local models) — Completely free and private, runs on your machine. Best for privacy-sensitive work, but requires a decent GPU for fast responses.

For your first setup, we recommend starting with a cloud model (OpenAI or Anthropic) to get the best experience. You can always switch to a local model later. The cost for casual use is typically under RM5 per month.

Step 4: Start OpenClaw and Send Your First Message

With Telegram and your AI model configured, start OpenClaw by running: npx openclaw start — in your terminal. You should see a log message confirming the Telegram connection is active. Now open Telegram, find your bot, and send a simple message like "What can you do?" OpenClaw will respond through the bot with a summary of its capabilities.

Try a few practical commands to see it in action: "What is the weather in Kuching?" or "Summarise the top news headlines today." These use OpenClaw's built-in web browsing skill to fetch real-time information and respond directly in your Telegram chat.

Step 5: Install Your First Skills

Skills are plugins that give OpenClaw new abilities. The default installation includes basic web browsing, file management, and conversation memory. To add more capabilities, use the skill installer:

Email: openclaw skill add email — connects to Gmail or Outlook for sending and reading emails.
Calendar: openclaw skill add calendar — manages Google Calendar or Outlook events.
Web Scraping: openclaw skill add scraper — extracts structured data from websites.
File Manager: openclaw skill add files — creates, edits, and organises documents.
Cron Jobs: openclaw skill add cron — schedules recurring tasks that run automatically.

Important: Only install skills from the official OpenClaw registry or trusted sources. A security audit found that third-party skills can execute code with full system access. Review what permissions each skill requests before installing it.

Step 6: Automate Your First Task

Here is a practical first automation that demonstrates OpenClaw's power. Send this message to your Telegram bot: "Every weekday at 8am, check the Ringgit to USD exchange rate and send me a summary." OpenClaw will set up a cron job that runs daily, fetches the exchange rate from the web, and messages you the result before your workday starts.

This single automation replaces a manual daily routine that takes 2-3 minutes each time — saving roughly 10 hours per year. Now imagine stacking dozens of these small automations: morning news briefings, invoice reminders, social media monitoring, meeting prep notes. Each one is a few minutes saved, but together they transform how you work.

Security Best Practices for Business Use

If you are using OpenClaw for anything beyond personal experimentation, follow these security guidelines. Both security researchers and the OpenClaw team recommend treating your agent like a new employee — give it only the access it needs.

Create separate service accounts for email, calendar, and other integrations. Never use your personal credentials.
Run OpenClaw in a sandboxed environment — a dedicated virtual machine or container, not your main workstation.
Audit every skill before installing. Check the source code, review permissions, and only use skills from the official registry.
Enable logging to track what OpenClaw does. Review logs weekly for unexpected actions.
Keep OpenClaw updated. Security patches are released frequently — the project had 512 vulnerabilities identified in its first audit.
Do not give OpenClaw access to financial systems, banking, or highly sensitive customer data without professional security review.

Common Beginner Mistakes to Avoid

After helping several clients explore OpenClaw, we have seen the same mistakes repeated:

Giving the agent your personal email and calendar access instead of creating dedicated service accounts.
Installing unverified third-party skills that request excessive permissions.
Running OpenClaw on a machine with access to production databases or financial systems.
Not monitoring the agent's actions — it can send emails and execute commands autonomously, so you need oversight.
Starting with complex multi-step workflows instead of mastering simple single-task automations first.

What to Do Next

Once you are comfortable with basic OpenClaw operation, the possibilities expand rapidly. Connect multiple messaging platforms so you can interact via WhatsApp or Slack. Build custom skills tailored to your business processes. Set up memory and context so OpenClaw remembers your preferences, clients, and ongoing projects. Explore the OpenClaw community on GitHub and Discord for skill sharing and troubleshooting.

For businesses that want to deploy OpenClaw at scale — with proper security, custom integrations, and enterprise-grade reliability — working with an experienced IT partner saves weeks of trial and error and avoids the security pitfalls that catch most DIY setups.

OpenClaw Beginner Guide: How to Set Up Your First AI Agent in 30 Minutes