Osaurus enables local and cloud AI models on Mac devices

As artificial intelligence models become more widely available and increasingly commoditised, many startups are now focusing on building the software layer that sits atop them. One of the latest companies entering that space is Osaurus, an open-source, Apple-focused LLM server that allows users to run and switch between local and cloud AI models while keeping their files, tools, and workflows on their own hardware.

Osaurus originated from an earlier desktop AI companion project called Dinoki. Osaurus co-founder Terence Pae described Dinoki as an “AI-powered Clippy,” inspired by Microsoft’s old assistant concept. However, customers using Dinoki repeatedly questioned why they needed to continue paying for AI tokens — usage-based credits charged by companies for processing prompts and generating responses — after purchasing the app. That feedback pushed Pae to explore running AI models locally rather than relying entirely on cloud providers.

“That’s how Osaurus started,” said Pae, who previously worked as a software engineer at Tesla and Netflix. He explained that the original idea was to build a locally running AI assistant capable of interacting directly with a user’s Mac system. “You can do pretty much everything on your Mac locally, like browsing your files, accessing your browser, accessing your system configurations,” Pae said. “I figured this would be a great way to position Osaurus as a personal AI for individuals.”

Pae began developing the software publicly as an open source project, continuously adding new features and addressing bugs as the platform evolved.

Today, Osaurus supports connections to both locally hosted AI models and cloud-based providers, including OpenAI and Anthropic. Users can choose which AI model to use while keeping important aspects of the experience — such as stored memory, files, and tools — on their own devices.

Flexibility is one of the platform’s main selling points, since different AI models often perform better on different tasks. Osaurus allows users to switch between models based on their needs at any given moment. The system functions as a “harness,” serving as a control layer that links AI models, workflows, and tools through a single interface. Similar platforms include OpenClaw and Hermes, though those tools are often designed primarily for developers comfortable using terminal-based interfaces.

Osaurus instead focuses on providing a more user-friendly experience for everyday users while also addressing security concerns. The company says the software operates inside a hardware-isolated virtual sandbox, which limits what the AI can access and helps protect user data and the host computer. Running AI models locally still requires powerful hardware, however. According to Pae, systems need at least 64GB of RAM to run local models effectively. Larger models, such as DeepSeek V4, may require systems with around 128GB of RAM.

Still, Pae believes local AI technology is rapidly improving. “I can see the potential of it, because the intelligence per wattage — which is like the metric for local AI — has been going up significantly,” he said. “Last year, local AI could barely finish sentences, but today it can actually run tools, write code, access your browser, and order stuff from Amazon. It’s just getting better and better.”

Osaurus currently supports a broad range of AI models, including MiniMax M2.5, Gemma 4, Qwen3.6, GPT-OSS, Llama, and DeepSeek V4. It also supports Apple’s own on-device foundation models and Liquid AI’s LFM family of on-device models. For cloud integrations, the platform can connect with OpenAI, Anthropic, Gemini, xAI’s Grok, Venice AI, OpenRouter, Ollama, and LM Studio.

The platform also works as a full MCP (Model Context Protocol) server, allowing compatible clients to access user tools and integrations. Osaurus ships with more than 20 built-in plug-ins for apps and services, including Mail, Calendar, Browser, Music, Git, Filesystem, Search, XLSX, PPTX, and more. More recently, the platform added voice support as well. Since launching nearly a year ago, Osaurus says the software has been downloaded more than 112,000 times. The company competes with other local AI tools such as Ollama, Msty, and LM Studio, but positions itself as a more approachable option for non-developers.

Terence Pae and Sam Yoo co-founded Osaurus, and the founders are currently participating in the New York-based startup accelerator Alliance. Looking ahead, the team is also considering business-focused use cases, particularly in industries like legal services and healthcare, where privacy concerns make locally hosted AI systems more attractive.

The founders believe that improvements in local AI could eventually reduce dependence on large cloud infrastructure and AI data centres. “We’re seeing this explosive growth in the AI space where cloud AI providers have to scale up using data centres and infrastructure, but we feel like people haven’t really seen the value of local AI yet,” Pae said.

“Instead of relying on the cloud, they can actually deploy a Mac Studio on-prem, and it should use substantially less power. You still have the capabilities of the cloud, but you will not be dependent on a data centre to be able to run that AI,” he added.