Osaurus: Open-source Apple Silicon LLM server brings local AI to Macs
Osaurus is an open-source, Apple Silicon–native language model server designed to run local LLMs on M1/M2 (and newer) Macs. Lightweight (around 7 MB), Swift-native, and compatible with the OpenAI REST API, Osaurus aims to be a performant Ollama alternative that keeps models and data on device for better privacy and offline use.
Key features
- Apple Silicon optimized (uses Apple ML frameworks) for improved speed on M1/M2/Pro/Max chips
- Extremely small footprint (~7 MB binary)
- Runs local models entirely offline — no cloud calls by default
- Implements the OpenAI-compatible REST API for easy integration
- Supports advanced features like KV-cache reuse, tool calling, and chat templates
Why it matters
For macOS users who prioritize data privacy and want low-latency local AI, Osaurus offers a compact, open-source option that leverages Apple Silicon performance. It can be useful for developers building local assistants, researchers testing models privately, or anyone who needs offline LLM capabilities.
Get it / Try it
Official repository and downloads: https://github.com/dinoki-ai/osaurus
Notes & comparisons
Osaurus is often compared to Ollama; its advantages are Apple-native optimizations and a minimal install size. As with any local LLM server, model availability and performance will depend on which model binaries you use.
Discussion
Would you run local models on your Mac? Share what use-cases you’d try (productivity tools, code assistants, research, or privacy-focused chatbots).
Post compiled from project repo and public coverage. No RSS links included.