🍺 pls - Blazingly Fast Context-Aware AI CLI

pls is a minimalist, blazingly fast Linux command-line utility written in C++. It translates natural language requests into ready-to-use bash/zsh commands using local LLMs via Ollama.

Unlike hundreds of bloated Python wrappers, pls runs instantly (zero-overhead), has no third-party dependencies, and is completely context-aware of your system environment.

IMG_7068.MOV

✨ Killer Features

• ⚡ Zero-Overhead: Written in raw C++ using libcurl. Manual JSON parsing ensures microsecond execution times.
• 🐧 OS Context-Aware: Automatically reads /etc/os-release to detect your Linux distribution and preferred package manager (zypper, apt, pacman, dnf).
• 👁️ Directory Context-Aware: The AI "sees" the files in your current directory. Asking to "delete log files" will generate a command specifically for the files in your folder.
• 🚰 UNIX Pipes (stdin) Support: Feed logs or errors directly into the AI: cat error.log | pls "find the issue and fix it".
• 📋 Wayland Native: Out-of-the-box clipboard integration via wl-copy.
• 🔒 100% Private: Everything runs locally on your hardware. No API keys, no data harvesting.

🛠️ Installation

Dependencies: libcurl and cmake. (e.g., for openSUSE: sudo zypper in libcurl-devel cmake gcc-c++)

git clone https://github.com/YOUR_USERNAME/pls-ai-cli.git
cd pls-ai-cli
mkdir build && cd build
cmake ..
cmake --build .
sudo cp pls /usr/local/bin/

🚀 Usage

First, tell pls which local Ollama model to use (we highly recommend qwen2.5-coder:3b for the best speed/intelligence ratio):

pls --set qwen2.5-coder:3b

(you can see ollama active models via pls --list)

Basic translation:

pls update my system

Output: zypper refresh && zypper dup (if you are on openSUSE)

Execution (-e) and Clipboard (-c) flags:

pls -e clear docker cache

Generates the command and asks: 🚀 Execute? [y/N]:

Pipe Magic (stdin):

ps aux | pls -e "find the command to kill the process consuming the most memory"

Output: kill -9 $(ps aux | sort -k4,4nr | head -n 1 | awk '{print $2}')

⚙️ Under the Hood

LLM generation temperature is hardcoded to 0.0 for maximum logical accuracy and zero hallucinations. Active model config is safely stored in ~/.config/pls_model.txt. Directory and Pipe context sizes are strictly limited to prevent memory overflow and VRAM crashes.