archdots

Minimal Arch Linux dotfiles for a local-first AI development setup using niri, DankMaterialShell, OpenCode, and llama.cpp.

This repository is intentionally small and sanitized. It does not include private hostnames, Wi-Fi names, tokens, SSH/GPG material, browser/editor state, shell history, KDE Connect keys, LM Studio state, or machine-specific mount points.

What This Includes

• A minimal .profile with local AI environment variables and useful aliases.
• A minimal Zsh setup with ZDOTDIR=$HOME/.config/zsh.
• A reusable niri config with DMS keybindings.
• A small DMS settings file with Catppuccin-inspired defaults.
• OpenCode config for a local llama.cpp OpenAI-compatible endpoint.
• Optional OpenRouter OpenCode config example with no API key included.
• Scripts to build llama.cpp with HIP and run llama-server.

What This Does Not Include

• Secrets or API keys.
• Personal shell history.
• Private machine names, Wi-Fi names, IPs, or account data.
• KDE, GNOME, browser, editor, or cloud-provider state.
• Full Arch installation automation.

Suggested Packages

Install only what you need. This is the package set I would start with on a fresh Arch machine:

paru -S niri dms-shell dms-shell-niri swayidle matugen wl-clipboard cliphist \
  xwayland-satellite xdg-desktop-portal-gtk xdg-desktop-portal-gnome \
  kitty zsh zsh-completions zsh-autosuggestions fzf zoxide fastfetch \
  git base-devel cmake ninja jq ripgrep fd eza bat btop yazi \
  nodejs npm rustup go python python-pip docker docker-compose \
  mesa vulkan-radeon libva-mesa-driver mesa-vdpau radeontop \
  rocm-hip-sdk rocm-opencl-sdk rocblas hipblaslt

You also need OpenCode. Follow the official install instructions from opencode.ai and keep the config from this repo under .config/opencode/opencode.json.

For AMD GPUs, check your target with:

rocminfo | rg -i 'gfx[0-9]+'

The llama.cpp build script defaults to gfx1151, which is the target used for the Radeon 8060S on the ASUS ROG Flow Z13. Change AMDGPU_TARGETS if your GPU is different.

Install Dotfiles

This repo is designed to be used with GNU Stow from the repository root.

git clone https://github.com/deepu105/archdots.git ~/archdots
cd ~/archdots
stow .

If you already have files in the same locations, review them first. Do not blindly overwrite a live setup.

Local Model Layout

The scripts use these default paths:

~/Workspace/llms/llama.cpp
~/Models

You can override them with environment variables:

export LLAMA_CPP_REPO="$HOME/Workspace/llms/llama.cpp"
export LLAMA_MODEL_ROOT="$HOME/Models"
export LLAMA_AMDGPU_TARGETS="gfx1151"

Put your GGUF models under ~/Models, for example:

~/Models/Qwen3.6-27B-GGUF/Qwen3.6-27B-Q4_K_M.gguf

Build llama.cpp

scripts/update_llama_cpp.sh

This clones or updates llama.cpp and builds llama-server and llama-bench with HIP support.

Start Local llama-server

llamaServer

Or run the script directly:

scripts/llama_server.sh

The server listens on:

http://127.0.0.1:18080/v1

OpenCode uses that endpoint from .config/opencode/opencode.json.

Optional OpenRouter

If you also want frontier models through OpenRouter, copy the example config and set your API key in your shell or secret manager:

export OPENROUTER_API_KEY="your-key-here"
cp .config/opencode/opencode.openrouter.example.json ~/.config/opencode/opencode.json

Do not commit real API keys.

Validate

niri validate -c .config/niri/config.kdl
bash -n scripts/llama_server.sh
bash -n scripts/update_llama_cpp.sh
zsh -n .profile
zsh -n .zshenv
zsh -n .config/zsh/.zshrc

Notes

This is a starting point, not a distro. Expect to adjust display names, scaling, GPU target, package list, and keybindings for your machine.