NanoBorealis: Agentic Linux Desktop with Free Models

What changed

NanoBorealis is presented as an agentic Linux desktop, built upon the Aurora distribution, which itself is a KDE Plasma desktop environment running on Fedora Atomic from the Universal Blue family. The core innovation is an integrated AI agent, referred to as 'nanobot,' designed to write, execute, and debug code directly on the user's machine. This agent operates within a secure sandbox, preventing unauthorized access to the user's files or network.

The system supports running on free cloud models via OpenRouter, with an automatic fallback mechanism if a primary model becomes unavailable. A key feature is the ability for users to contribute their own hardware, including GPUs, to the agent's processing power. This is managed through the NanoBorealis app, which benchmarks available hardware to select the most suitable model. For larger models, multiple computers can be pooled together using the nanoborealis pool command.

NanoBorealis is designed with user control and security in mind. The agent runs rootless in its own account, behind a firewall, and without elevated OS privileges. Chat history is synchronized across desktop, WebUI, and the NanoBorealis app, with all data stored locally on the user's machine. Updates are managed manually, with builds tested and promoted through development, testing, and stable channels, ensuring users can revert to previous versions if needed.

Installation can be done via a dedicated NanoBorealis app for Windows, macOS, and Linux, which can create a bootable USB stick pre-configured with an OpenRouter key. Alternatively, users can download the installer manually and write it to a USB stick. Existing Aurora users can switch to NanoBorealis without a full reinstallation using a bootc switch command.

The project is built as a bootc image, meaning the entire operating system is a single, signed container image updated daily from the latest Aurora build. Each build undergoes rigorous testing, including internal image checks and agent/app test suites. Two editions are available: a standard version and a NVIDIA edition that includes NVIDIA's open driver and container toolkit for GPU acceleration. The project utilizes a three-branch system (dev, testing, main) for managing OS and app updates, with images signed using cosign.

Why it matters for builders

For AI builders, NanoBorealis offers a unique platform to develop and test agentic applications in a controlled, local environment. The ability to run agents with sudo privileges within a sandboxed Linux system, coupled with direct access to hardware resources and the use of free models, provides a powerful and cost-effective development setup. This architecture allows for experimentation with agent autonomy and resource management without the typical cloud infrastructure costs or complexities.

Practical impact

Builders can explore creating agents that interact with a Linux environment, install tools, build projects, and manage code directly on their local machine. The pooling feature enables testing of larger models that might not fit on a single GPU. The project's emphasis on local data storage and manual updates also provides a blueprint for building more secure and user-controlled AI applications. Developers can start by installing the NanoBorealis app and creating an install stick, or by switching an existing Aurora installation.

Caveats and source limits

The provided source material indicates that GPU pooling is a planned feature, currently running on processors, which may limit performance for demanding agent tasks. The roadmap also mentions upcoming features like a login screen and welcome app, suggesting the user interface is still evolving. While the project is licensed under Apache 2.0, specific details regarding commercial use or enterprise-level support are not elaborated upon in the provided excerpts. The source also notes that the project was formerly known as nanobot OS and NanoAurora, with systems set up under those names migrating automatically.