Osaurus
osaurus is an Apple Silicon-native local LLM server built on Apple's MLX framework. It exposes OpenAI-compatible endpoints via a SwiftNIO server and includes a SwiftUI desktop app. By targeting the MLX framework — optimised for the M-series chip's unified memory architecture — it achieves faster inference and lower power consumption than general-purpose local LLM servers on Apple hardware.
Ollama is the established standard for local LLM serving and supports a broad range of models and platforms. osaurus occupies a narrower position: it runs only on Apple Silicon and supports a smaller model set, but trades breadth for performance on that specific hardware. The MLX backend takes advantage of the M-chip's unified memory in ways that Metal-based backends in Ollama do not fully exploit.
Placed in Assess at outer position because the performance advantage is real but the constraints — Apple Silicon only, narrow model support, smaller community and ecosystem — limit applicability. Worth evaluating for developers doing sustained local inference work on M-series hardware who have already tried Ollama and found its performance ceiling. The outer position reflects that most developers will be adequately served by Ollama; osaurus is a specialist option for a specific hardware context.
