Running Your Own AI Agent

Step 1: Install Linux

Arch provides a minimal, customizable base. No bloat, no mystery processes—just what you need. Sway is a tiling Wayland compositor that keeps the desktop lightweight and keyboard-driven. Combined with LY as the login manager, you get a fast, distraction-free environment that stays out of your way.

This stack is ideal for developers who live in terminals and want their AI agent running on dedicated, efficient hardware while still providing you a Window manager and essential tools for working on and with your AI assistant.

Base Installation

Start with the archinstall script for a clean Arch base. Select Sway as your desktop environment and LY as the display manager. After installation, chroot into the system and update your mirror list for faster package downloads:

sudo reflector --verbose --sort rate -l 20 --save /etc/pacman.d/mirrorlist

Essential Packages

Install the core toolkit for development and daily use. This includes Thunar for file management with archive and media plugins, Starship for a customizable prompt, and various utilities for screenshots, fonts, and system tools. Wezterm provides a modern terminal, while rustup prepares the system for Rust-based tools.

Don't forget base-devel for compiling AUR packages and git for version control.

AUR and Browser Setup

After rebooting, install paru from the AUR for easier package management. Then add Google Chrome for web testing and development. Set up your local bin directory and configure Starship in your bashrc for a better terminal experience.

Development Tools

Install Fresh Editor for terminal-based text editing and LazyGit for streamlined Git workflows. These tools integrate seamlessly with the terminal-centric workflow and keep you productive without leaving the keyboard and keeping system resources free for your agent.

Intall steps

• Install Arch using the archinstall script.
• Sway for desktop (lightweight and fast)
• LY for login manager
• Chroot into system after install and install the following:
• sudo pacman -S reflector
• sudo reflector --verbose --sort rate -l 20 --save /etc/pacman.d/mirrorlist
• sudo pacman -Syu thunar thunar-archive-plugin thunar-media-tags-plugin thunar-volman tumbler ffmpegthumbnailer gvfs gvfs-smb gvfs-mtp starship zed ttf-firacode-nerd php fzf fuzzel nwg-panel nwg-shell nwg-look-bin wezterm grim slurp swappy noto-fonts-emoji rustup bind rdap less xdg-desktop-portal xdg-desktop-portal-gtk xdg-desktop-portal-wlr
• sudo pacman -S –needed base-devel git
• Reboot and finish setup

After reboot

• sudo pacman -Syyu
• Install paru
• git clone https://aur.archlinux.org/paru.git
• cd paru/
• makepkg -si
• paru -S google-chrome
• mkdir ~/.local/bin
• echo ‘eval "$(starship init bash)"’ >> ~/.bashrc
• echo 'export PATH=$HOME/.local/bin:$PATH' >> ~/.bashrc
• echo ‘export PATH=$HOME/none_modules/bin:$PATH’ >> ~/.bashrc

Extras

• Install Fresh Editor: - curl https://raw.githubusercontent.com/sinelaw/fresh/refs/heads/master/scripts/install.sh | sh
• Install LazyGit - sudo pacman -S lazygit
• Install dd_dotstore - vailable at https://bitbucket.org/ldnddev/dd_dotstore

Step 2: Install Prerequisites

Before installing OpenClaw, prepare your service accounts. Create a dedicated Google account for API access, set up a Google Cloud project, and enable Gmail and Drive APIs. Download the client_secret.json for gogcli authentication.

Acquire a dedicated phone number—using an old device works well for separation. Set up Telegram with a new bot via @BotFather for OpenClaw's messaging interface.

Create accounts on OpenRouter for LLM access and Brave Search for web search capabilities. These provide the brains and research capabilities for your agent.

• Create new Google account (Will be used durring OpenClaw installation)
• Setup new project in google cloud
  • Enable api services: https://console.cloud.google.com/apis/dashboard
  • Add credentials desktop app: https://console.cloud.google.com/apis/credentials
  • Add Gmail and Drive APIs
  • Download client_secret.json (store this in a safe place will be used durring GogCLI installation)
• Install gogcli: https://github.com/steipete/gogcli
• Authorize new gmail account using client-secret.json
• We used an old cellphone and acquire new phone number for that device. (You could use your own but we preferr a level of seperation)
• Setup new Telegram account with new number
  • Setup new bot chanel @BotFather (Will be used durring OpenClaw installation)
• Setup OpenRouter account - https://openrouter.ai/keys (Will be used durring OpenClaw installation)
• Setup Brave Search for API key - https://brave.com/search/api/ (Will be used durring OpenClaw installation)
• Install openclcaw
• Add telegram
• Set heartbeat to only run once per hour and use a cheaper model than main model (will save $)

Step 3: Configure OpenClaw

OpenClaw is a lightweight, feature-rich AI agent that runs locally on your (it's) own hardware. It connects to your chat platforms, manages your files, schedules tasks, controls browsers, and remembers everything across sessions. Best part? It can run on a $100 refurbished Lenovo ThinkCentre M75q-1 Tiny. No cloud required. No monthly fees. And No, you don't need an expensive Apple machine.

1. Install OpenClaw - curl -fsSL https://openclaw.ai/install.sh | bash
2. Install and use Node 25 nvm install 25 nvm use 25 node --version # Should show v25.x.x
3. Install qmd: npm install -g @tobilu/qmd
4. Add to PATH (in ~/.bashrc): export PATH="$HOME/.npm-global/bin:$PATH"
5. Create collection: qmd collection add ~/.openclaw/workspace/memory --name openclaw-memory --mask "**/*.md"
6. Generate embeddings: cd ~/.openclaw/workspace/memory qmd embed

Environment Variables

Add these to ~/.bashrc or ~/.zshrc:

# gog default account export GOG_ACCOUNT=YOUR_NEW_GMAIL_ACCOUNT # Bun (available but qmd uses npm/Node 25) export PATH="$HOME/.bun/bin:$PATH" # qmd (REQUIRED - npm global bin) export PATH="$HOME/.npm-global/bin:$PATH" # Local bin export PATH="$HOME/.local/bin:$PATH" # NVM (for Node version switching) export NVM_DIR="$HOME/.nvm" [ -s "$NVM_DIR/nvm.sh" ] && \. "$NVM_DIR/nvm.sh" # Use Node 25 by default (for qmd/agent compatibility) nvm use 25 &>/dev/null || true

Quick Health Check Commands

# Verify OpenClaw curl -s http://127.0.0.1:18789/status | head -20 # Verify gog GOG_ACCOUNT=YOUR_NEW_GMAIL_ACCOUNT gog auth list # Verify Node (should be v25) node --version nvm current # Verify qmd qmd --version qmd status qmd search "test" --json # Verify Rust cargo --version # Verify Git git --version # Check Bun bun --version

OpenClaw QMD skill

• npx playbooks add skill openclaw/skills --skill qmd-cli

What You Get

For that $100 ThinkCentre, you now have: web search and browsing, file management, task scheduling via cron, memory across sessions, multiple model support through OpenRouter, and extensible skills via markdown files in `~/.openclaw/skills/`.

The Real Costs

Running your own agent isn't free. OpenRouter API calls run $5-20/month. Electricity is under $10/year. But your time is the real cost—initial setup takes 3-4 hours. You're your own DevOps team. No redundancy means power outages take you offline.

The Alternative: KiloClaw

Maybe you like OpenClaw but hate becoming a part-time sysadmin. KiloClaw is OpenClaw, hosted. Same agent, but someone else handles infrastructure. One-click deploy, auto-restart, 500+ models, team management, and automatic updates. Pricing starts at $49/month with Kilo Pass Pro.

Bottom Line

OpenClaw democratizes AI agents. For under $100 in hardware and an afternoon of setup, you're running the same tool developers pay cloud providers hundreds to access. Yes, there are tradeoffs. But there's something satisfying about owning your infrastructure. Your data stays local. Your agent runs on your terms.

Until next time, Jared Lyvers