top of page
Search

Install OpenClaw: The AI that actually does things.

  • Writer: Aastha Thakker
    Aastha Thakker
  • 6 hours ago
  • 5 min read

If you’re building with AI tools, you’ve heard the buzz around agents. But why is everyone trying to run their own AI assistant when Claude.ai and ChatGPT already exist?


Because those tools are viewers. You open a tab, you type, you close the tab. OpenClaw (also called Clawd Bot) changes that entirely. Instead of you typing to the AI, the AI works for you.


This difference “passive chatbot vs. active AI agent” is what makes this project special. And yes, it does sound a bit wild when you first hear it.


What Is OpenClaw (Clawd Bot), Really?


You’ll bump into this project under several names: Clawdbot, Moltbot, Clawd Bot, OpenClaw. Here’s the short version of why:


The project was built by Peter Steinberger. He originally called it Clawdbot. Anthropic’s lawyers flagged it as too close to their “Claude” trademark, so it became Moltbot, and eventually landed on OpenClaw, clean, memorable and open source.


OpenClaw isn’t an AI, it’s a gateway. Think of it as a persistent middleman that sits on your machine and connects your chat apps to your AI provider.


Here’s the simple workflow:

WhatsApp / Telegram / Slack / Discord / Google Chat / Signal / iMessage / BlueBubbles / Microsoft Teams / Matrix / Zalo / Zalo Personal / WebChat
               │
               ▼
┌───────────────────────────────┐
│            Gateway            │
│       (control plane)         │
│     ws://127.0.0.1:18789      │
└──────────────┬────────────────┘
               │
               ├─ Pi agent (RPC)
               ├─ CLI (openclaw …)
               ├─ WebChat UI
               ├─ macOS app
               └─ iOS / Android nodes

Let’s set this up


You don’t need to be a senior engineer to follow this guide. If you’ve run commands in a terminal before and created an account on a developer platform, you’re good.


Hardware

  • Any machine running macOS, Linux, or Windows with WSL2

  • At least 4GB RAM

  • Stable internet connection (every message you send makes an API call)

  • The machine should stay on while you’re using the bot, if it sleeps, the bot goes dark


Accounts to Create

  • Telegram Account

  • Anthropic or Google account


Step 1: Install OpenClaw from the Official Site

  1. Open your browser and go to the official OpenClaw site: OpenClaw — Personal AI Assistant or Moltbot.

2. Scroll down to the QuickStart section. You’ll see a one-line install command. Copy it.

3. Open your terminal and paste and run that command:

4. The installer will pull down all dependencies automatically. You’ll see packages being downloaded. Let it finish. Once done, the OpenClaw banner will appear in your terminal, and the setup wizard will launch automatically.


Step 2: Configure Your AI Provider

  1. When asked for your AI provider, select Google AI Studio (or Anthropic, your choice). I have used Google AI Studio for API here.

  2. Head to aistudio.google.com, grab your API key, and paste it in when asked. Make sure to check the usage limits on your selected model before committing.

  3. Select your preferred model and hit Next.

4. The wizard will ask about optional extras, skills, memory settings, extras. You can safely hit No on all of these for now. You can always configure them later.


Step 3: Create & Connect Your Telegram Bot

1. Open Telegram and search for @BotFather. Make sure it has the blue verified checkmark; that’s the official one.

2. Start a chat with BotFather and send “/newbot”. BotFather will ask you things. Name: the display name users see and Username: which must end with bot.

3. BotFather will reply with your bot token, a long string. Treat this token like a password. Anyone who has it can send and receive messages as your bot. Keep it out of git repos and public config files.

4. Open your newly created bot in Telegram and hit “/start”. It will show message like this. Use this code when it is asked while configuring. Your bot will respond with a pairing code, a short code. This is how OpenClaw verifies it’s really you connecting.

5. Back in the terminal wizard, when it asks for the pairing code, paste in what your bot sent you. Hit enter.


Security Considerations:


Running OpenClaw isn’t just installing a chatbot. You’re deploying a local agent that can execute actions on your machine. That changes the threat model entirely.

  1. Do not run OpenClaw as root or administrator. If the AI executes a malicious or hallucinated command, the damage will be limited to the current user’s permissions.

  2. Protect Your Tokens Like Passwords. You now have two sensitive secrets: Telegram Bot Token and AI Provider API Key. If either leaks, attackers can impersonate your bot, they can burn your API credits or they can send malicious instructions to your agent. Never commit them to Git, never store them in public config files or share screenshots containing them. If exposed, rotate immediately.

  3. Every Telegram message triggers an API call. If your bot is public anyone can spam it, your API bill can spike quickly. To mitigate it keep bot private unless needed to be public, use rate limits and monitor usage dashboard frequently.

  4. Even if it’s ‘just personal automation,’ pause and ask: What data can it access? What commands can it run? What happens if it’s hijacked? What happens if it hallucinates? If you can’t answer those confidently, tighten your configuration before using it heavily.


Step 4: OPENING THE WEB UI & YOUR FIRST CHAT

1. When the wizard asks, choose Open the Web UI.

2. Your browser will open at http://localhost:18789. This is the OpenClaw control panel.

3. Your bot is now live. Head back to Telegram, open your bot’s chat, and say hello.


Some features you can see on the left panel are:


While you’re in the Web UI, here’s what each item in the left panel has, worth knowing before you start exploring yourself.

  1. Usage is your billing dashboard. Every time you send a message, OpenClaw makes an API call to your AI provider and that costs tokens.

  2. Cron Jobs is where you set up anything you want your AI to do automatically, on a timer, without you having to ask.

  3. Agents lets you create more than one AI assistant, each with its own name, system prompt, model, and set of allowed tools. Think of each Agent as a separate employee with a specific job description. Each Agent can be assigned to a different channel too, so your Work Agent only talks in your work Telegram group, and your Personal Agent only talks in your private DM. They’re completely separate.

  4. Config is the visual editor for everything in your `~/.openclaw/openclaw.json` file. Instead of editing raw JSON in a terminal, you can change your API keys, switch models, update channel settings, adjust memory limits, and manage permissions — all from a form interface in your browser.

  5. Logs is the persistent, scrollable history of all gateway activity — every message received, every API call made, every Cron Job that fired, every Skill that executed. It shows a clean event-level view that’s easier to read and search.


OpenClaw is not dangerous. Unrestricted automation is.


The difference between a powerful AI assistant and a liability is configuration discipline.


Set boundaries first. Automate second.


Now stop reading and give it something real to do!!

Comments

bottom of page