I've been running OpenClaw for a while — connecting it to Telegram, WhatsApp, and Slack as my daily AI assistant. Here's a step-by-step walkthrough to get it up and running in under 30 minutes.
Step 1: Install OpenClaw
The fastest way — one command:
curl -fsSL https://openclaw.ai/install.sh | bash
Or if you prefer npm:
npm install -g openclaw@latest
openclaw onboard
Requirements: Node.js 22+ (Windows users need WSL2).
The onboard command walks you through initial setup — API key, model selection, and your first channel connection. The whole process takes about 5 minutes.
You can also install via pnpm, or git clone the repo if you want to hack on it. There's even a macOS companion app in beta.
Step 2: Pick Your AI Model
OpenClaw is model-agnostic. Here are your options:
| Model | Best For | Cost |
|---|---|---|
| Claude Sonnet | Best balance of speed & quality | ~$3/1M tokens |
| Claude Haiku | Budget-friendly, fast responses | ~$0.25/1M tokens |
| GPT-4o | Alternative if you prefer OpenAI | ~$2.50/1M tokens |
| Ollama (local) | Free, fully offline | $0 (your hardware) |
For most people, I'd recommend starting with Claude Haiku — it's cheap, fast, and surprisingly capable for daily tasks.
Step 3: Connect Your First Channel
This is where it gets fun. Let's connect Telegram as an example (it's the easiest):
- Message
@BotFatheron Telegram - Send
/newbotand follow the prompts - Copy the bot token
- Add it to your OpenClaw config
# ~/.openclaw/config.yaml
channels:
telegram:
bot_token: "YOUR_BOT_TOKEN"
allowed_users:
- your_telegram_id
- Restart OpenClaw — done. Message your bot and it responds with AI.
Want WhatsApp instead? It's a bit more involved — you need the WhatsApp Business API. Set up a Meta Business account, create a WhatsApp app, and configure the webhook URL pointing to your OpenClaw instance. The trickiest part is getting the permanent access token, but the official Meta docs walk you through it.
Discord? Also straightforward — create a bot in the Discord Developer Portal, enable the Message Content intent, set up proper permissions (Send Messages, Read Message History, etc.), and add the bot token to your config. The setup is similar to Telegram.
OpenClaw supports 50+ channels in total, including Slack, Microsoft Teams, Google Chat, LINE, Matrix, and more. Each has different setup complexity — Telegram and Discord are the easiest to start with.
Step 4: Add Skills (This Is the Fun Part)
Skills are what turn OpenClaw from a chatbot into an actual agent. There are 60+ curated skills available.
Some favorites:
- GitHub CLI — Create PRs, review code, manage issues from chat
- Linear — Create and update tickets without opening a browser
- Home Assistant — Control smart home devices via natural language
- Tavily Web Search — Real-time web search with AI summarization
- Docker Essentials — Manage containers from your messaging app
Each skill has a safety badge:
- Recommended — Well-tested, low risk
- Caution — Involves sensitive operations (like file deletion)
- Experimental — Newer, needs community testing
You can browse and install skills through the CLI with openclaw skills list and openclaw skills add <name>.
Step 5: Expose It Securely (Optional)
If you want to access OpenClaw from outside your local network:
Don't open ports directly. Instead, use one of these:
- Tailscale — Zero-config VPN, my recommended approach
- Cloudflare Tunnel — Free, no port forwarding needed
- Nginx reverse proxy + Let's Encrypt — Classic but more setup
The key principle: never expose your OpenClaw instance directly to the internet without authentication and encryption.
What Can You Actually Do With It?
Here's what a typical day looks like with OpenClaw running:
Morning (via WhatsApp):
"What's on my calendar today?"
"Summarize the 3 GitHub issues assigned to me"
During work (via Slack):
"Create a Linear ticket: fix the auth bug in the login flow, priority high"
"Review PR #142 and leave comments"
Evening (via Telegram):
"Set living room lights to 30%"
"What's the weather tomorrow?"
It remembers your preferences, your conversation history, and your routines. It's not a stateless chatbot — it's a persistent agent.
Cost Breakdown
Running this costs less than you'd think:
| Component | Cost |
|---|---|
| OpenClaw | Free (MIT license) |
| Hosting | Free (your own machine) |
| AI Model (Haiku, moderate use) | ~$5-10/month |
| Channels (Telegram, Discord) | Free |
| Total | ~$5-10/month |
Compare that to $20/month for ChatGPT Plus, which only works in a browser tab.
Wrapping Up
Self-hosting an AI agent sounds intimidating, but OpenClaw makes it genuinely accessible. The install takes minutes, the model costs are low, and having AI in your existing messaging apps — instead of yet another browser tab — is a surprisingly big quality-of-life upgrade.
Give it a try. If you get stuck, open-claw.me has pretty solid installation guides, channel comparisons, and a skills directory that I've found useful.
Have questions or want to share your setup? Drop a comment below — I'd love to see what integrations people are using.
Top comments (0)