Imagine having an AI assistant that runs 24/7, responds to you on Telegram, WhatsApp, Slack, or Discord, executes tasks on your server, and costs less than a coffee subscription to host. That's exactly what OpenClaw delivers — and setting it up is simpler than you'd think.
OpenClaw is the most starred open-source AI assistant project on GitHub (354,000+ stars). It's self-hosted, privacy-first, and connects to virtually every messaging platform you already use. In this guide, we'll walk through setting it up on a VPS with Claude as the AI brain, turning your cheap Linux server into a powerful personal assistant.
What Is OpenClaw and Why Should You Care?
OpenClaw is a TypeScript-based AI assistant platform that runs a Gateway (a WebSocket control plane) on your own hardware. The Gateway orchestrates everything: the AI model, your messaging channels, browser automation, file operations, scheduled tasks, and more.
What makes it different from ChatGPT or Claude's web interface:
- Always-on: Runs as a daemon on your server — no browser tab needed
- Multi-channel: One brain, 24+ messaging platforms (Telegram, WhatsApp, Slack, Discord, iMessage, Signal, and more)
- Self-hosted: Your data stays on your server. No conversation logs sent to third parties
- Model-agnostic: Works with Claude, GPT-4, Gemini, DeepSeek, Grok, local models via Ollama — or all of them with automatic failover
- Extensible: 5,400+ community skills on ClawHub for coding, research, automation, and more
What You'll Need
Before we start, here's the shopping list:
- A VPS: Any Linux server with 1GB+ RAM. Hetzner, DigitalOcean, or Contabo work great — $4-10/month is enough
- Node.js 22.16+ (Node 24 LTS recommended)
- An AI provider API key: Anthropic (Claude), OpenAI, or any supported provider
- A Telegram bot token (or whichever channel you want to use first)
Step 1: Install OpenClaw on Your VPS
SSH into your server and run the global install:
npm install -g openclaw@latestThen launch the interactive onboarding wizard:
openclaw onboard --install-daemonThe onboard command walks you through everything: Gateway configuration, workspace setup, channel connections, and skill installation. The --install-daemon flag creates a systemd service so OpenClaw starts automatically on boot and survives SSH disconnects.
Alternatively, if you prefer Docker:
git clone https://github.com/openclaw/openclaw.git
cd openclaw
docker compose up -dStep 2: Connect Claude as the AI Brain
OpenClaw supports multiple LLM providers. To use Claude (our recommendation for coding and reasoning tasks), you need an Anthropic API key.
During the onboard wizard, select Claude/Anthropic as your provider and paste your API key. If you've already completed setup, update the configuration:
openclaw config set ai.provider anthropic
openclaw config set ai.apiKey sk-ant-your-key-here
openclaw config set ai.model claude-sonnet-4-20250514OpenClaw also supports model failover — you can configure multiple providers so if one is down or rate-limited, it automatically switches to another. This is powerful for a 24/7 assistant that can't afford downtime:
openclaw config set ai.fallback.provider openai
openclaw config set ai.fallback.model gpt-4oStep 3: Set Up Telegram as Your Primary Channel
Telegram is the most popular channel for OpenClaw — and the easiest to set up. Here's how:
- Open Telegram and message @BotFather
- Send
/newbotand follow the prompts to create your bot - Copy the bot token (looks like
1234567890:ABCdefGhIJKlmNoPQRsTUVwxYZ) - In your VPS terminal, connect the channel:
openclaw channel add telegram
# Paste your bot token when promptedNow message your bot on Telegram. The first message triggers a DM pairing request. Approve it from your server:
openclaw pairing approveThat's it — you now have a personal AI assistant in your Telegram. Every message you send goes to Claude (or your configured model), and the response comes back instantly.
Step 4: Add More Communication Channels
The real power of OpenClaw is one brain, many interfaces. You can add any of these channels with the same openclaw channel add command:
- WhatsApp — via the Baileys library (no Business API needed)
- Slack — via Bolt, perfect for work contexts
- Discord — via discord.js, great for community management
- iMessage — via BlueBubbles (requires a Mac somewhere in the chain)
- Signal — via signal-cli, for the privacy-conscious
- Google Chat, Microsoft Teams, Matrix, IRC, LINE, Feishu (Lark) — and more
Each channel can be configured independently — different personalities, different allowed commands, different security levels. You might want your Telegram bot to have full server access while your Discord bot only answers questions.
What Tasks Can Your OpenClaw Assistant Handle?
Once set up, your assistant isn't just a chatbot. It's a capable agent that can:
Daily Productivity
- Set reminders and scheduled tasks via built-in cron jobs
- Summarize long articles, documents, or email threads you paste in
- Draft emails, messages, and social media posts
- Translate between languages in real-time
- Track habits, journal entries, or daily logs
Development and DevOps
- Execute shell commands on your VPS
- Monitor server health and alert you when something breaks
- Run browser automation tasks (scraping, testing, form filling) via built-in CDP control
- Interact with GitHub — review PRs, check CI status, create issues
- Query databases and APIs on your behalf
Research and Knowledge
- Search the web and summarize findings
- Analyze documents, CSVs, and data files on your server
- Monitor RSS feeds, news, or competitor websites on a schedule
- Build a personal knowledge base using skills like memU
Home and Life Automation
- Control smart home devices (with appropriate integrations)
- Track expenses by messaging receipts to your bot
- Get weather, traffic, or calendar briefings on a schedule
- Process photos and files sent via messaging apps
Securing Your Assistant
Since your assistant has server access, security matters. OpenClaw has sensible defaults:
- Gateway binds to loopback only — not exposed to the internet by default
- DM pairing — unknown senders get a pairing code you must approve manually
- Allowlists — restrict which Telegram/Discord users can interact with the bot
- Docker sandboxing — run untrusted sessions in isolated containers
- Tailscale integration — access the web dashboard securely without exposing ports
For a personal assistant, the DM pairing approach is perfect: only you (and people you explicitly approve) can talk to your bot.
Remote Access: Connecting Your Devices
Even though the Gateway runs on a VPS, you can pair companion apps on your local devices:
- macOS menu bar app — quick access from your desktop
- iOS and Android apps — mobile control with voice support
- CLI access —
openclawcommands from any machine via SSH tunnel
Device nodes can execute local actions (camera, screen recording, notifications, clipboard) while the AI brain runs on your server. This gives you the best of both worlds: always-on cloud processing with local device capabilities.
Cost Breakdown: What Does This Actually Cost?
One of OpenClaw's biggest advantages is cost efficiency:
- VPS: $4-10/month (Hetzner CX22 or DigitalOcean basic droplet)
- AI API: $5-20/month for moderate personal use (Claude Sonnet is very cost-efficient)
- Messaging channels: Free (Telegram, Discord, Slack bots are free to create)
- Total: $10-30/month for a 24/7 personal AI assistant across all your devices
Compare that to $20/month for ChatGPT Plus (single user, single interface, no automation) or $20/month for Claude Pro (same limitations). With OpenClaw, you get more flexibility for less money — and full control over your data.
Frequently Asked Questions
Can I use OpenClaw without a VPS?
Yes. OpenClaw runs on any machine — your laptop, a Raspberry Pi, or a Mac Mini. A VPS is recommended for 24/7 uptime, but it's not required. You can also run it on Cloudflare Workers using the community moltworker project.
Does OpenClaw store my conversations?
Conversations are stored locally on your server by default. Nothing is sent to OpenClaw's servers — there are no OpenClaw servers. It's fully self-hosted. The only external calls are to your configured AI provider (Claude, OpenAI, etc.) for inference.
Can I use multiple AI models at the same time?
Yes. OpenClaw supports model failover (automatic switching when one provider is down) and multi-agent routing (different channels or users can be routed to different models). You could use Claude for coding questions and GPT-4o for creative writing, for example.
Is OpenClaw safe to expose to the internet?
The Gateway binds to loopback (127.0.0.1) by default and is not internet-accessible. Messaging channels use outbound connections only — your bot connects to Telegram's servers, not the other way around. For web dashboard access, use Tailscale Serve or an SSH tunnel rather than exposing ports directly.
How does OpenClaw compare to running Claude Code CLI directly?
Claude Code CLI is a developer tool focused on coding tasks in a terminal. OpenClaw is a general-purpose assistant platform that can use Claude Code as one of its engines (via the openclaw-claude-code plugin). OpenClaw adds messaging channels, scheduled tasks, voice, browser control, and a skills ecosystem on top. Think of it as Claude Code for your entire digital life, not just your codebase.
Getting Started Today
The barrier to running your own AI assistant has never been lower. A $5 VPS, a free Telegram bot, and 10 minutes of setup gives you a private, always-on assistant that works across every platform you use. OpenClaw's onboarding wizard handles the complexity — you just answer a few questions and start chatting.
If you're a developer, the extensibility is where it gets really exciting. The ClawHub skill registry has 5,400+ community skills, MCP server support via the mcporter bridge, and a plugin architecture that lets you build custom integrations for anything. Your assistant grows with your needs.
The future of AI isn't locked behind a $20/month subscription to a single provider's web interface. It's self-hosted, multi-channel, and under your control. OpenClaw makes that future accessible today.