Building a Personal AI Agent with OpenClaw

What is OpenClaw?

OpenClaw is an open-source AI agent platform that runs 24/7 on your own hardware. Think of it as a personal assistant that never sleeps — connected to your messaging apps, armed with real tools, and fully under your control.

Unlike cloud-based AI chatbots, OpenClaw connects to platforms like Telegram, Discord, and WhatsAppas its interface. It has access to a real shell, a browser, the filesystem, cron scheduling, email, web search, and more. And because it's self-hosted, your data never leaves your machine.

It acts as a genuine personal assistant — with persistent memory across sessions, scheduling capabilities, web research, and the ability to take real actions on your behalf.

The Setup

The whole thing runs on a Dell OptiPlex micro PCI had sitting in a closet. It's small, quiet, sips power, and runs Ubuntu Server 24/7.

The stack is straightforward:

• Hardware: Dell OptiPlex Micro (refurbished, ~$150)
• OS: Ubuntu Server with Node.js v24
• Runtime: OpenClaw gateway, always connected
• Interface: Telegram bot for chatting from anywhere
• Access: Cloudflare tunnel for external connectivity

The bot is connected to my Gmail for reading and sending emails, has browser automation for web tasks, shell access for server management, and can run scheduled cron jobs. It's like having a junior developer on call 24/7.

Architecture

Here's how everything fits together — a hub-and-spoke model with the OpenClaw gateway at the center:

What Can It Do?

Here's a sampling of tasks I've delegated to the agent:

• Email management — Read incoming Gmail, draft replies, send messages. I get a summary of important emails throughout the day.
• Portfolio deployments — Push code to GitHub and the agent handles building and deploying to Vercel. This very blog post was deployed by the agent.
• Domain & DNS management — Manage DNS records via the Cloudflare API, set up tunnels for external access.
• Scheduled tasks & reminders — Cron jobs for periodic checks, reminders at specific times, heartbeat polls for routine maintenance.
• Web research — Search the web, fetch articles, summarize findings. Much faster than doing it manually.
• Code execution — Run shell commands, manage the server, install packages, debug issues. Full terminal access from Telegram.
• Persistent memory — Notes, context, and decisions persist across sessions. It remembers what we discussed yesterday.

Why Self-Host?

You might wonder — why not just use ChatGPT or Claude directly? There are a few reasons I prefer self-hosting:

• Privacy.My emails, files, shell history — none of it leaves my server. With cloud services, you're trusting a third party with everything.
• Cost. The LLM providers I use (z.ai, Groq) have generous free tiers. The hardware was a one-time ~$150 investment. No monthly subscription.
• Control. Full access to the filesystem, shell, browser — the agent can do anything I can do at the terminal. No sandboxing restrictions.
• Always on.It's running 24/7. I can send a message at 3 AM and get a response. Cron jobs run on schedule regardless.

The whole setup — from bare metal to a working agent — took me about an hour. The real magic starts when you customize it to your own workflow.