Hey Product Hunt community! I built IronMule out of a straightforward need: Apple Silicon's unified memory architecture is incredible for local LLMs, but getting transparent, reproducible benchmarks and true hardware-native inference shouldn't require bloated setups or opaque tooling. What is IronMule? It’s a lightweight, open-source runtime and benchmarking tool built specifically for Apple Silicon using Apple’s MLX framework. It helps you get the most out of your Mac when running models locally—measuring real throughput, prompt-processing speed, and interactive token latency without overhead. Why MLX? Instead of generic cross-platform wrappers, IronMule taps directly into Apple’s native unified memory and Metal execution, allowing for fast, transparent local testing and comparison across different quantization levels and model architectures. The project is 100% open source on GitHub. I’d love to get your feedback, hear what models you're currently running locally on your machines, and see how IronMule performs on your hardware! Thanks for checking it out!

IronMule

High-performance local LLM runtime for Apple Silicon

High-performance local LLM runtime for Apple Silicon