I wanted a place where an AI coding session could keep its context instead of starting from a blank terminal every time.
That is the idea behind PI-Desktop: a local-first desktop workspace for AI coding agents. It brings persistent projects, sessions, model switching, permissions, and multi-agent workflows into one app that runs on macOS, Windows, and Linux.
The current release is v0.15.1. The repository has roughly 4.6k GitHub stars and 390 forks.
What I use it for
- Persistent projects and sessions — project context and conversation history stay available across restarts.
- Model freedom — switch providers and models without rebuilding the whole workflow around one vendor.
- Plan and Goal modes — move from a vague request to an editable plan, then execute it with visible progress.
- Permission controls and diff review — keep destructive actions behind approval and review file changes before accepting them.
- Plugins and MCP — extend the workspace with tools and integrations instead of putting every workflow into the core app.
- Multi-agent workflows — delegate independent work to subagents and keep the main session focused.
Why a desktop workspace?
Terminal coding agents are powerful, but long-running work quickly turns into a collection of terminals, scrollback, copied prompts, and half-remembered decisions. A desktop workspace gives those sessions a home.
PI-Desktop is built with Electron, a Rust host core, and the pi agent harness. The goal is not to hide the agent behind a simplified chat box. It is to make the surrounding work easier to inspect: what the agent changed, which tools it used, what still needs approval, and where another agent can help.
The local-first approach also matters to me. Your projects and sessions should remain usable without handing the whole workflow to a hosted dashboard.
Links
If you build with AI coding agents, I would especially like feedback on the session model, permissions, and multi-agent workflows.
Top comments (0)