Local LLM Playground

Local LLM Playground is an AI playground for running local language models directly on your iPhone.

The app supports both GGUF and MLX language models, and now also supports Apple Foundation Models. From entering prompts to viewing inference results, you can easily experiment with on-device AI entirely on your device.

The following AI models are bundled with the app, so you can start using local AI immediately after installation:

• Qwen2.5-0.5B-Instruct-Q4_K_M (GGUF)
• Qwen3-0.6B-MLX-4bit (MLX)

No cloud connection or API key is required. All AI processing is performed locally on your device, and your prompts and personal data are never transmitted to external servers. In addition to prompt experimentation and model evaluation, the app provides detailed inference metrics, including:

• Token generation speed (tokens/sec)
• Inference time (prefill / decode)
• Token usage statistics
• Memory usage
• Context length (n_ctx)

You can also import your own GGUF and MLX models to experiment with a wide variety of local AI models. Key Features:

• Load GGUF language models
• Load MLX language models
• Support for Apple Foundation Models
• Run AI inference from custom prompts
• Display token generation speed
• Visualize inference timing
• Monitor memory usage
• Configure context size
• Advanced inference parameter settings
• Import GGUF models
• Import MLX models

Highlights:

• Fully on-device AI
• Supports GGUF, MLX, and Apple Foundation Models
• No cloud services required
• No API keys required
• Privacy-focused by design
• Ideal for experimenting with and learning about local AI

Recommended For

• Anyone interested in local LLMs
• Developers exploring MLX models
• Users who want to experience Apple Foundation Models
• AI enthusiasts evaluating model behavior
• Prompt engineering and experimentation
• Comparing the performance of lightweight language models

Important Notice: AI-generated responses may contain inaccuracies or incorrect information. Please independently verify any important information before relying on it.