OpenClaw (formerly ClawdBot) has taken the tech world by storm — and for good reason. It’s an open-source, self-hosted personal AI assistant that runs on your own hardware, connects to messaging apps you already use, and actually takes real actions on your behalf.
It manages your email. Books your calendar. Checks you in for flights. Controls your smart home. Browses the web. Runs terminal commands. And it does all of this through a conversation in WhatsApp, Telegram, Slack, Discord, Signal, or iMessage.
But setting it up properly is where most people hit a wall.
What You Need
Hardware. OpenClaw runs locally on your machine. A Mac Mini is the most popular choice — reliable, energy-efficient, and powerful enough to run 24/7. But it works on any Mac, PC, or Linux server.
An AI model API key. OpenClaw works with Claude (recommended), ChatGPT, Gemini, and local models via Ollama. Most people use Claude for the best results.
Messaging channel access. WhatsApp requires the WhatsApp Business API or a connected phone. Telegram needs a bot token. Slack, Discord, and others have their own setup requirements. Each channel has its quirks.
The Installation
The basic installation is straightforward — brew install clawdbot or npm i -g clawdbot, then clawdbot onboard to run the setup wizard. The wizard walks you through gateway configuration, workspace setup, channel connections, and initial skill installation.
That gets you a working assistant. But a working assistant and a useful assistant are different things.
What Separates a Good Setup from a Great One
Custom skills are where OpenClaw becomes truly powerful. Out of the box, it can chat. With custom skills, it can manage your specific email workflow, interact with your specific tools, and automate your specific routines. Building skills that match your actual life and work is where the real value lives.
Proactive routines transform OpenClaw from reactive to proactive. Instead of waiting for you to ask, it sends morning briefings, monitors for specific triggers, runs scheduled tasks, and alerts you when something needs attention. This is configured through cron jobs and scheduled skills.
Voice integration via ElevenLabs gives your assistant a natural speaking voice. Wake word activation on macOS means you can talk to it hands-free. Combined with the messaging channels, you can interact with your assistant however is most convenient in the moment.
Memory and context are what make OpenClaw feel personal. It stores your preferences, past conversations, and ongoing context as local files. The more you use it, the better it gets. But the initial memory setup — telling it about your preferences, your tools, your routines — dramatically accelerates this.
Common Pitfalls
Skipping the onboarding conversation. The onboarding skill exists for a reason. It helps OpenClaw learn about you — your name, preferences, tools, routines. Skip this and your assistant starts from zero every conversation.
Not configuring security properly. OpenClaw has access to your machine. Sandboxed execution, access controls, and trust boundaries exist for a reason. Configure them thoughtfully.
Trying to do everything at once. Start with one or two channels and a handful of skills. Get those working well, then expand. The community has hundreds of skills available — you don’t need to install them all on day one.
Neglecting the daemon. Running clawdbot start manually means your assistant dies when you close the terminal. The daemon (--install-daemon) keeps it running in the background, surviving restarts, and always available.
Who Should Use OpenClaw?
OpenClaw is for people who want full control over their AI assistant — what it can access, where it runs, how it behaves. It’s for tinkerers, builders, and anyone who’s frustrated by the limitations of cloud-based assistants.
It’s not a plug-and-play consumer product (yet). It requires some technical comfort to set up and maintain. But for those willing to invest the setup time, it’s the closest thing to having a real digital assistant that exists today.