Osaurus brings both local and cloud AI models to your Mac

In a rapidly evolving landscape of artificial intelligence, startups are scrambling to create innovative software layers that enhance user experience. One notable newcomer is Osaurus, an open-source LLM (Large Language Model) server designed exclusively for Apple devices. This platform allows users to seamlessly switch between local AI models and cloud solutions, all while keeping their files and tools securely on their own hardware. Osaurus emerged from the vision of a desktop AI assistant, initially dubbed Dinoki, which co-founder Terence Pae described as a modern-day 'AI-powered Clippy.' Feedback from Dinoki users prompted Pae to rethink the necessity of token payments for AI processing. This led him to explore the potential of running AI locally on devices. "That’s how Osaurus started," Pae, who previously worked as a software engineer at Tesla and Netflix, explained in a recent interview. The primary goal of Osaurus is to facilitate a personal AI experience directly on users’ Macs. With capabilities like file browsing, web access, and system configuration management, this tool positions itself as a valuable companion for individuals. Since its inception as an open-source project, Osaurus has evolved with new features and bug fixes, ultimately enabling connections to both locally hosted AI models and cloud services such as OpenAI and Anthropic. Users can easily select their preferred AI models while maintaining control over their data and tools. The flexibility of this system allows users to switch between AI models based on their specific needs, making Osaurus a unique 'harness' for various AI tools and workflows. Unlike other platforms focused on developers, Osaurus is designed with a user-friendly interface, addressing security concerns by operating within a hardware-isolated virtual sandbox. This setup ensures that the AI remains within a limited scope, safeguarding users' computers and data. Although running AI models locally requires significant resources, with a minimum of 64 GB of RAM recommended, Pae believes that as technology advances, these requirements will decrease. "The intelligence per wattage for local AI is on an upward trend," he observed. Osaurus is capable of running several models, including MiniMax M2.5, Gemma 4, and GPT-OSS, among others. Additionally, it connects to Apple’s on-device models and a variety of cloud services. Recently, Osaurus has introduced voice capabilities, enhancing its functionality further. Since its launch nearly a year ago, the tool has garnered over 112,000 downloads. As the founders, including Sam Yoo, participate in the New York-based startup accelerator Alliance, they are exploring future applications for Osaurus, particularly in sectors like law and healthcare, where local LLMs could significantly address privacy concerns. With the potential for local AI models to reduce reliance on expansive data centers, Pae highlights a shift in the AI landscape. "Instead of depending on cloud solutions, users can deploy a Mac Studio on-premises, using considerably less power while still benefiting from cloud-like capabilities," he stated, underscoring the transformative possibilities of local AI.