PI-Desktop is a local-first desktop application that provides a workspace for AI coding agents, built with Electron, Rust, and TypeScript.
The tool addresses the fragmentation of AI coding workflows across terminals, editor extensions, and hosted services by offering a unified desktop environment where agents can work on projects while remaining under user control. It operates on a local-first model with no mandatory account or relay service, allowing users to bring their own language model from any OpenAI-compatible provider or local source. The architecture uses a Rust host core with the pi Agent Harness and supports user-installable plugins that can contribute tools, commands, panels, themes, and services. Work flows through three modes: Agent mode for autonomous execution, Plan mode for reviewing a frozen implementation before execution, and Goal mode for approving outcomes after the agent determines its approach. All privileged actions pass through a permission layer where users can inspect diffs, review command output, and control session autonomy.
Developers should adopt this tool if they want agent-assisted coding without editor lock-in or dependency on hosted services. It suits projects where maintaining visibility and control over agent actions matters more than pure automation speed. The tool distinguishes itself by operating entirely locally with no mandatory relay, supporting multiple concurrent models and providers, and offering a dedicated workspace rather than embedding agents within existing development tools.
The project is actively developed and already usable for real coding workflows, though APIs and extension interfaces remain subject to evolution. Development activity shows ongoing work across the desktop application, plugin system, and agent capabilities, with support for internationalization and multiple language models integrated into the core design.
