penguin

Simple, minimalist AI-Agent for Linux that runs on almost every device

LLM

Penguin uses (at the moment) the qwen3.5:2b model with Ollama as inference. The model can run at a good conversation speed on most modern laptops and a bit slower on older machines, the minimum I reached with an eight years old Lenovo Thinkpad(i5, 7th generation?) was 7 t/s. I recommend using an 32k context window, which is enough for most applications you will run with penguin.

Installing & Quick Start

One-line install command using pipx:

pipx is a modern PyPi Installer for CLI applications - like apt but for python. Use it to install the penguin-agent-package with a systemwide command penguin :

sudo pipx install penguin-agent-package

Then type penguin in your terminal

Standard install using pip:

pip is the standard PyPi Installer. Install penguin-agent-package using pip in a virtual environment:

pip install penguin-agent-package

[!WARNING] This will not install the systemwide penguin command. You need to add it manually to PATH.

Features

Penguin is a very, very simple tool:

• Uses an Ollama backend(see LLM)
• Three basic tools in the agent-loop:
  • execute_bash_command
  • get_memory_entry
  • enter_memory_entry
• Executes bash commands always at ~
• Uses prompt_toolkit to manage sessions with history and suggestions
• Uses memory.json file for managing the agent memory

Run it

Just type:

penguin

How it works

The choice of Ollama is due entirely to its good tool-calling integration in python. Tool schemes are converted by the ollama library to JSON and passed to the native function calling of the model.

Credits

I heavily used the prompt_toolkit library for the design and the interaction with the tool. Thanks also to the amazing ollama python library that makes it so easy to use LLMs with Python.