LiteRT Skill for Claude Code

A Claude Code skill for Google LiteRT — the universal framework for on-device AI (successor to TensorFlow Lite).

Installation

npx skills add farmhutsoftwareteam/litert-skill

Or manually clone to your skills directory:

git clone https://github.com/farmhutsoftwareteam/litert-skill ~/.claude/skills/litert

What This Skill Provides

This skill gives Claude Code expertise in deploying ML and GenAI models on edge devices using LiteRT.

Coverage

Topic	Description
GPU Acceleration	OpenCL, OpenGL, Metal, WebGPU setup with zero-copy buffers
NPU Acceleration	Qualcomm & MediaTek support, AOT/JIT compilation
LiteRT-LM	Deploy LLMs (Gemma, Phi, Qwen) on device
Model Conversion	PyTorch, TensorFlow, JAX → .tflite
Migration	TensorFlow Lite → LiteRT migration guide

Supported Platforms

Platform	CPU	GPU	NPU
Android	✅	✅	✅
iOS	✅	✅	Soon
macOS	✅	✅	Soon
Windows	✅	✅	Soon
Linux	✅	✅	—
Web	✅	✅	Soon

When This Skill Triggers

• Working with .tflite models
• On-device ML/AI inference
• Mobile ML deployment (Android/iOS)
• GPU/NPU hardware acceleration
• Running models like Gemma on device
• Migrating from TensorFlow Lite

Skill Structure

litert-skill/
├── SKILL.md                    # Main skill file
└── references/
    ├── gpu-acceleration.md     # GPU setup & optimization
    ├── npu-acceleration.md     # NPU setup & vendor support
    ├── litert-lm.md            # LLM deployment
    ├── model-conversion.md     # Model conversion guides
    └── migration.md            # TFLite migration

Resources

• LiteRT Documentation
• LiteRT GitHub
• LiteRT-LM GitHub

License

MIT