Magnitude is an open-source inference engine, licensed under Apache 2.0, designed to optimize the execution of large language models on consumer-grade hardware. The software performs automated hardware profiling to estimate performance metrics, specifically tokens per second (tok/s), for various models and quantizations before they are downloaded. This allows users to select models based on a ranking system that considers speed, accuracy, intelligence, and memory requirements. Technically, Magnitude optimizes model execution through end-to-end tuning tailored to specific hardware configurations, implementing advanced techniques such as speculative decoding and context size adjustments. The engine supports a wide array of hardware architectures, including Apple Silicon, NVIDIA GPUs, AMD GPUs, and CPU-only setups, with specific compatibility for unified-memory systems like Strix Halo and DGX Spark. It features an automated memory management system that loads models on demand and unloads them when idle or when memory limits are reached to ensure stability. Users can integrate the engine with various AI agents—including Pi, OpenCode, Hermes, and Claude Code—via a one-click connection process. Distributed as a native desktop application for macOS, Windows, and Linux, Magnitude also includes an integrated CLI, providing a private, offline environment that eliminates API costs and rate limits.