Pi Desktop
A local-first AI coding workspace for macOS, Windows and Linux that supports local or cloud models, parallel agents and workspace plugins.
Pi Desktop is an open source desktop workspace for developers who want AI coding agents outside a specific editor or terminal. It keeps projects, sessions, code reviews and previews in a persistent environment on macOS, Windows and Linux. You can connect local models, cloud providers or custom gateways and switch models within a session.
The agent can read code, edit files, run commands and test its work. Plan mode lets you review an approach before execution; Goal mode gives the agent an objective and acceptance criteria for longer tasks. Privileged operations pass through a permission layer.
For parallel work, subagents handle independent tasks in their own contexts and report back to the parent. Worker Sessions handle longer assignments, with each worker available as a separate session you can inspect. Sessions also support branching, search and queued prompts, alongside diff reviews and command output.
Plugins extend both the agent and the desktop. They can add skills, agent tools, panels, floating widgets and background services, or connect local and remote MCP servers. The application combines Electron, a Rust host core and the pi Agent Harness, and uses the LGPL-3.0 license.
No account is required. Projects, sessions, settings and logs stay local by default, credentials live in the OS keychain, and Pi Desktop collects no telemetry. With a remote model, the context needed for a request goes directly to your configured provider without a mandatory Pi Desktop relay.
