@turtle170/rift

Rift — Your desktop pet code reviewer powered by local AI. 🦀 🦞

Rift is an interactive, local-first code analysis companion. It runs directly on your machine using llama.cpp to parse, analyze, and review your code repositories. It features a rich terminal user interface (TUI) with a customizable pet identity, live animations, and parallel code parsing.

Features

• 🥚 Interactive Hatching: Run rift hatch to dynamically generate your companion pet's name, personality traits, and stats based on your machine GUID.
• 📂 Multi-Language Tree-sitter Support: Parses Rust, Python, JavaScript, TypeScript, C, C++, Go, and Java.
• ⚡ Tokio/Rayon Parallel Analysis: Spawns concurrent file workers (called Claws) to process your codebase's AST into streamlined Markdown S-expressions.
• 💻 Automatic GPU Acceleration: Automatically detects your system's hardware configuration (WMI / dedicated VRAM check) and fetches either the Vulkan-optimized or CPU-only build of llama.cpp.
• 🌶️ Roast Mode: Let your pet critique your code with absolutely zero empathy and maximum pedantry.
• ❓ Grill Mode: Interrogation mode where the pet asks you probing, difficult questions about your architecture choices instead of just listing bugs.

Installation

You can install Rift globally using npm:

npm install -g @turtle170/rift

Note: Rift is currently optimized for Windows systems (x64).

Quick Start

1. Hatch Your Pet

To initialize Rift, download the local LLM model (~6.4 GB), and configure your pet:

rift hatch

2. Analyze a Project

Review your source code files under any directory:

rift analyze ./src

3. Roast Mode

Instruct your pet to review your code with harsh, pedantic critiques:

rift roast ./src

4. Grill Mode

Instruct your pet to interrogate you and ask questions about your code layout:

rift grill ./src

Options

• --max-files <LIMIT>: Set the maximum number of files to inspect during an analysis (defaults to 50 to prevent context overflow).

License

This project is licensed under the Apache License 2.0.