pulseaiclub/phi

What it solves

It provides a lightweight, high-performance terminal-based interface for interacting with LLMs to perform coding tasks. It addresses the bloat and latency often found in GUI-based AI coding assistants by offering a minimal, fast, and keyboard-driven environment that runs directly in your terminal.

How it works

It operates as a terminal user interface (TUI) that connects to various LLM providers (OpenAI, Anthropic, Gemini). It gives the model a set of core tools—such as file reading, writing, editing, and bash command execution—to complete tasks. It features a permission gate for safety, supports sub-agents for complex workflows, and allows for extensibility through a binary protocol (PXB) and Model Context Protocol (MCP) servers.

Who it’s for

Developers who prefer working within the terminal and require a fast, resource-efficient AI assistant that can handle complex coding tasks, git operations, and file management without the overhead of heavy IDE extensions or Electron-based applications.

Highlights

• Extremely lightweight: ~15 MB binary with very low idle RAM usage.
• High performance: ~31 ms time-to-first-frame for instant responsiveness.
• Native git integration: Includes a full-screen diff review tool within the terminal.
• Flexible operation: Supports both an interactive TUI and a headless mode for automated tasks or CI/CD pipelines.
• Secure: Built-in permission system to gate destructive actions like file deletions or shell commands.