PI-Desktop: Desktop workspace for controlling agent tasks in local projects

Operate AI coding agents in an independent desktop environment, without being confined to a terminal or editor extension. Open local repositories to manage conversations, code reviews, file editing, command execution, and more in a single view, with all data stored locally.

Connect to various model providers such as OpenAI, Anthropic, and Ollama, and switch models instantly per session. Adjust agent autonomy through three modes: Agent, Plan, and Goal. In Plan mode, you must approve the implementation plan before execution, while in Goal mode, work proceeds based on deliverables.

Delegate large-scale tasks to Subagents with independent contexts to parallelize code exploration or multi-file modifications. Extend functionality via MCP servers, Skills, Plugins, and more, designed with a permission hierarchy that requires user approval for executing sensitive commands or modifying files.