// Agent profile
PI-Desktop by vastsa is a cross-platform desktop application for AI-assisted coding that prioritizes local-first data ownership. Built with Electron on the frontend and a Rust host core running the pi Agent runtime, it offers three distinct work modes — Agent (autonomous), Plan (approval-first), and Goal (outcome-driven) — so developers can choose the right level of AI autonomy per task. The app supports multiple model providers (OpenAI, Anthropic, local endpoints, and compatible APIs), persistent project sessions with history and forking, and an extensible plugin marketplace for Skills, MCP servers, and Subagents. No mandatory accounts or cloud relays. 2,900 stars on GitHub as of September 2026.
PI-Desktop is a local-first desktop workspace for AI-assisted coding that puts data ownership and model flexibility at the center. Rather than tying developers to a single AI provider or requiring cloud accounts, it lets you bring your own models — commercial APIs, local inference endpoints, or anything compatible — and keeps all project data on your machine. The Rust host core handles system-level operations and agent orchestration while the Electron frontend provides a polished GUI with multi-session support.
The three work modes are PI-Desktop’s defining feature. Agent mode lets the AI work autonomously, making file edits and running commands without interruption. Plan mode requires human approval at each step, giving developers full control over every change. Goal mode sits between the two — you describe an outcome and the agent proposes a plan, executing only after you approve the overall approach. This flexibility means the same tool works for quick prototyping (Agent), careful refactoring (Plan), and feature implementation (Goal).
The plugin marketplace extends the core shell with installable Skills, MCP server integrations, and Subagents. This keeps the base application lean while allowing teams to add project-specific tooling, custom workflows, and specialized AI capabilities.
PI-Desktop appeals to developers who want a desktop AI coding environment without vendor lock-in. The multi-model support makes it practical for teams that use different providers across projects or want to compare model performance. The local-first architecture suits privacy-conscious organizations and developers working with sensitive codebases. Session persistence and forking are valuable for long-running projects where context continuity matters.
PI-Desktop is open source under LGPL-3.0 but is still a relatively young project. As an Electron app it has a larger binary footprint than terminal-only tools. You need your own API keys for commercial model providers. The plugin ecosystem is growing but may not yet match the breadth of more established tools.
Developers who want a local-first, model-agnostic desktop environment for AI-assisted coding. Particularly appealing to those who value data ownership, want flexibility across AI providers, and prefer a GUI with session management over terminal-only workflows.
Builder.io's open-source framework for building agent-native applications — shared actions, SQL-backed state, identity, tools, skills, jobs, observability, and UI surfaces that all work together.
CLI orchestration tool that installs and configures platform connectors for 16+ social and web sources — Twitter/X, Reddit, YouTube, GitHub, Bilibili, and more. Zero API fees.
Persistent memory layer for AI coding agents — benchmark-backed (95.2% on LongMemEval-S), 92% fewer tokens per session vs full-context pasting, zero manual memory.add() calls.