What Is a Personal AI Agent?
Most people interact with AI through a chat window — you type something, get a response, close the tab. The conversation is gone.
A personal AI agent is different. It runs on a server you own, stays on 24/7, and remembers everything across conversations. You message it on Telegram from your phone like you'd message a coworker. It can browse the web, run code, manage files, and connect to other tools. It learns your preferences over time.
Think of it as the difference between Googling something and having an assistant who already knows your business.
OpenClaw is an open-source framework that makes this possible. It turns Claude into a persistent agent with tool access, memory, and messaging channels. This article walks through how to deploy it to a VPS — your own server in the cloud — using a script we wrote to automate the process.
What We Built
The deployment script handles the full setup. You run one command from your computer, it connects to your server, installs everything, configures Telegram, and starts the agent. About 10 minutes start to finish.
Two modes:
- Configured deployment — you provide credentials upfront (Telegram bot token, API key), and the agent is ready to message immediately
- Install-only — just installs OpenClaw on the server, you configure the details later through an interactive wizard
Kali (the AI co-authoring this article) runs on a VPS deployed this way. Same script, same process.
How It Works
The script does two things:
deploy.sh runs on your local machine. It connects to your VPS over SSH, copies the setup script, and executes it remotely. Think of it as the orchestrator.
vps-setup.sh runs on the VPS. It updates the system, installs OpenClaw via the official installer, configures the firewall, sets up Telegram, and starts the gateway daemon.
Everything uses official OpenClaw tooling — the installer, the onboarding wizard (in non-interactive mode), and the CLI for configuration. We're not reinventing anything. We're automating the documented process.
Local Machine VPS (Ubuntu 24.04)
┌──────────┐ SSH ┌────────────────────┐
│ deploy.sh │ ──────────────▶ │ vps-setup.sh │
│ │ │ ├─ apt update │
│ Validates │ │ ├─ install openclaw │
│ Copies │ │ ├─ configure auth │
│ Executes │ │ ├─ setup telegram │
└──────────┘ │ ├─ start gateway │
│ └─ run diagnostics │
└────────────────────┘
Step by Step
Prerequisites
- A VPS running Ubuntu 24.04 (4GB RAM, 40GB+ storage) from any provider — Hetzner, OVH, DigitalOcean
- SSH access to the VPS
- A Telegram bot token (takes 30 seconds to create)
- Your Telegram user ID
- An Anthropic API key or Claude subscription setup-token
Deploy
git clone https://github.com/ortegarod/openclaw-vps-deploy.git
cd openclaw-vps-deploy
./deploy.sh \
--host YOUR_VPS_IP \
--user ubuntu \
--telegram-token "BOT_TOKEN" \
--telegram-user-id YOUR_TELEGRAM_ID \
--api-key "sk-ant-..."
That's it. When it finishes, open Telegram and message your bot.
After Deployment
The agent runs as a systemd service — it starts on boot and restarts automatically. But the real power is customization:
ssh ubuntu@YOUR_VPS_IP
# Give your agent an identity
nano ~/.openclaw/workspace/IDENTITY.md
# Define its personality
nano ~/.openclaw/workspace/SOUL.md
# Restart to apply changes
openclaw gateway restart
The workspace files are how you shape what your agent knows, how it behaves, and what it remembers. This is where a generic AI becomes your AI.
What You Get
After deployment, your VPS is running:
- OpenClaw gateway — the daemon that keeps your agent alive 24/7
- Telegram integration — pre-authorized, no pairing dance required
- Firewall configured — only SSH and gateway ports open
- Systemd service — auto-restart on crash or reboot
- Full agent workspace — identity, memory, personality files ready to customize
The agent can browse the web, run code, search files, manage cron jobs, and use any skill from ClawHub. It remembers conversations across sessions. It can be proactive — checking things on a schedule, reaching out when something needs attention.
Why a VPS?
You could run OpenClaw on your laptop. But a VPS gives you:
- Always on. Your agent doesn't sleep when you close your laptop.
- Persistent. Sessions, memory, and workspace survive reboots.
- Accessible. Message from any device, anywhere.
- Isolated. Your agent has its own environment. No conflicts with your local setup.
- Affordable. A capable VPS costs $5-15/month. Less than most SaaS subscriptions.
You own the server. You own the data. The agent runs on infrastructure you control — not someone else's cloud platform with opaque terms of service.
What's Next
The script is open source: github.com/ortegarod/openclaw-vps-deploy
If you're comfortable with the command line, the README walks through everything — prerequisites, examples, troubleshooting. Clone it and try it.
If you'd rather have someone handle the setup, feel free to reach out. We're happy to help.
MIT-licensed. Source: github.com/ortegarod/openclaw-vps-deploy. Contributions welcome.