minghinmatthewlam/pi-gui

What it solves

pi-gui provides a graphical user interface for the pi coding agent, eliminating the need to interact with the agent solely via a command-line interface. It allows developers to manage multiple agent tasks in parallel, review code changes visually before committing them, and access essential development tools like terminals and file explorers within a single window.

How it works

It is an Electron-based desktop application that acts as a shell around the @earendil-works/pi-coding-agent. Rather than reimplementing the agent's logic, it uses a thin adapter (pi-sdk-driver) to communicate with the pi runtime. The app reads and writes session files (JSONL transcripts) directly from the disk, ensuring that settings, credentials, and threads are shared between the GUI and the pi CLI.

Who it’s for

Developers who use the pi coding agent and prefer a visual environment for managing complex coding tasks, reviewing diffs, and running tests side-by-side with agent conversations.

Highlights

• Parallel Agent Threads: Run multiple tasks simultaneously, each optionally in its own git worktree, with a sidebar for tracking status.
• Integrated Workbench: Combines a real terminal, file explorer, editor, and a dedicated Review tab for staging changes file-by-file.
• Thread Management: Ability to fork threads from any message, rewind sessions using /tree, and schedule recurring tasks.
• Agent Orchestration: Supports "threads that run threads," where one agent can initiate and communicate with other agent threads.
• Flexible Provider Support: Compatible with various model providers via OAuth, API keys, or custom endpoints.