The desktop is the agent - a first look at Praxis Linux (Phase 0)
90-second demo. Everything in it ran on a local model. The shape of the problem The desktop hasn't changed shape in thirty years. One application to look at files, another to write words, another to write code. We translate intent into a thousand mechanical motions - clicks, drags, menu paths and we stop noticing. When large language models arrived, the industry's answer was to bolt them onto the…
In a 90-second demo, Praxis Linux, a Linux distribution where the desktop is the agent, showcases its capabilities. The desktop has not changed shape in thirty years, with one application for each task such as files, text editing, and coding. However, with the arrival of large language models, the industry's response was to attach these AI assistants to the side, rather than integrating them into the core functionality.
Praxis Linux aims to change this approach by making the desktop the agent, with one surface - the Canvas - that listens, understands, and adapts to the user's needs.
The Canvas, built using Tauri 2 and React 19, is a declarative generative UI that composes from a vetted component catalog, ensuring security, accessibility, and consistency. It listens to the user's intent, switches to the appropriate mode (file explorer, editor, document, or settings panel), and performs the required action. For instance, when a user clicks on a file, the Canvas displays a one-paragraph summary of its contents and purpose.
The agent is powered by a Python-based router that handles routing between local and cloud models, memory management using Qdrant, SQLite, and DuckDB, and model execution through Ollama, a local-fast route. The user has full control over which models to use and can opt-in to cloud models when necessary. Every action taken by the agent is legible, reversible, and consented to, ensuring trustworthiness and user sovereignty.
Phase 0 of Praxis Linux is a pre-alpha spike, with no ISO available for installation yet. The project currently supports three modes: Conversation, File, and Code. Future phases will expand the functionality to include Document, Settings, and Browser modes. The project is open-source under the Apache 2.0 license, allowing users to bring their own models and ensuring no vendor lock-in. Contributions are welcomed through issues, discussions, and pull requests.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
