Osaurus launches open-source AI app combining local and cloud models on Mac

A new startup called Osaurus has launched an open-source LLM server for Mac that lets users seamlessly switch between local and cloud-based AI models while keeping all their files, memory, and tools on their own hardware. The app represents a growing movement toward privacy-first AI tools that give users control over their data without sacrificing the quality of cloud-based models.

Osaurus evolved from an earlier project called Dinoki, described by co-founder Terence Pae as an «AI-powered Clippy.» The pivot happened when Dinoki's customers kept asking why they needed to pay for both the app and AI tokens separately. That question led Pae — a former software engineer at Tesla and Netflix — to rethink the relationship between the AI interface and the AI backend.

The result is a Mac-native app that serves as a universal interface for AI models. Users can run powerful open-source models like Llama and Mistral locally using Apple Silicon's GPU, and seamlessly switch to cloud models like GPT-4 or Claude when they need more computational power. The key innovation is that regardless of which model is processing the request, all files, conversation memory, and tool configurations stay on the user's Mac.

This hybrid approach addresses a growing concern among professionals and developers who want to use AI tools without sending sensitive documents and code to cloud servers. With Apple Silicon Macs now offering enough GPU power to run capable models locally, the technical foundation for this approach has only recently become viable.

The open-source nature of Osaurus also means developers can extend and customize the tool for specific workflows — from code review to content writing to data analysis. As the AI model landscape continues to commoditize, the real value increasingly lies in the software layer that orchestrates these models, and Osaurus is positioning itself as that layer for the Mac ecosystem.