PI Desktop Open source

What it does

PI Desktop is a local-first desktop application, built on Electron with a Rust host core and a "pi Agent Harness," that gives AI coding agents a persistent workspace instead of a disposable terminal session. Work is organized as a Project → Session → Agent hierarchy rather than a single throwaway chat, so context, decisions and files stay attached to the project they belong to.

What is inside

• Three operating modes: Agent (direct read/edit/run execution), Plan (the agent proposes a plan before touching anything) and Goal (you state an outcome and the agent chooses its own path)
• Subagent delegation and a Session Orchestrator that coordinates multiple Worker Sessions in parallel
• Diff review, command inspection and work recovery, plus session branching and transcript navigation
• A plugin system covering agent tools, skills, completion handlers, workspace panels, floating widgets, themes, MCP servers and background services, with starter templates (panel-basic, agent-tool-basic, skill-pack, full-demo)

Pricing

Free and open source under the GNU Lesser General Public License v3.0. There is no paid tier, no mandatory account and no relay service — projects, sessions and settings stay on your machine.

Works with

Model-agnostic: OpenAI, Anthropic, any OpenAI-compatible API, Ollama, LM Studio, local models or a custom gateway, and you can swap models mid-session or per task without rebuilding anything.

How to install or connect

Pre-built packages for macOS (Apple Silicon and Intel), Windows x64 and Linux x64 are on the GitHub releases page. To build from source you need Node.js ≥22.19, pnpm ≥10 and a stable Rust toolchain:

git clone https://github.com/vastsa/PI-Desktop.git
cd PI-Desktop
pnpm install
cargo build -p host-core
pnpm build:js
pnpm dev

Maintenance and safety

LGPL-3.0 licensed, with 5.9k+ GitHub stars and active commits. A dedicated permission layer sits between agent and tool — every privileged action flows through "Agent → Tool Request → Permission Layer → Allow/Ask/Deny → Execution" — and you configure how much autonomy each session gets.

Who should use it

Developers who want a persistent, local-first home base for agent work across several projects at once, explicit control over how autonomous an agent session is allowed to be, and the option to extend the workspace themselves through its plugin API.