Osaurus launches Mac LLM server

Osaurus released an open-source macOS AI harness on May 15 that lets users run local language models on Apple Silicon and connect to cloud providers when they need more compute, according to the company’s GitHub repository and documentation. The software is positioned as a control layer for models, memory, tools and identity that stays on the user’s Mac rather than on a vendor’s servers. The repository describes the project as MIT-licensed, built in Swift and able to work fully offline with local models. A May 15 report by TechCrunch said the product is Apple-only and was developed by co-founder Terence Pae. ### What exactly did Osaurus ship on May 15? The GitHub repository for osaurus-ai/osaurus says Osaurus is “the AI harness for macOS” and sits between the user and “any model — local or cloud.” The project page says it provides agents with persistent memory, autonomous execution, real-code execution and remote reachability, while keeping the surrounding layer on the machine. (github.com) The documentation page lists release 0.18.13, 53,100 downloads and a MIT license. The same page says Osaurus supports local models, Apple Foundation models and cloud models during onboarding, and presents the app as a native macOS product for Apple Silicon. ### How does it handle local versus cloud models? Osaurus documentation says the product works fully offline with local models and can connect to cloud providers when a user wants more power. (github.com) The GitHub page says “nothing leaves your Mac unless you choose,” framing cloud access as optional rather than required. TechCrunch reported on May 15 that Osaurus can connect with locally hosted models as well as providers including OpenAI and Anthropic. (docs.osaurus.ai) The article said users can switch models while keeping memory, files and tools on their own hardware, which Pae described as part of a personal AI system built around the Mac. (github.com) ### Who is behind the project? TechCrunch identified Terence Pae as an Osaurus co-founder and said he previously worked as a software engineer at Tesla and Netflix. Pae told TechCrunch the idea grew out of Dinoki, a desktop AI companion, after customers questioned paying for both the app and model tokens. An older GitHub repository under dinoki-ai says the Osaurus project was moved to osaurus-ai/osaurus and that active development, issues and releases now live there. (techcrunch.com) That ties the current repository to the earlier Dinoki branding referenced in outside coverage. ### What does the software include beyond model routing? Osaurus Docs says the app includes a chat overlay, agent-specific prompts and histories, memory that distills prior conversations into compact facts, working folders with scoped file and git tools, and schedules or watchers for timed or folder-triggered runs. (techcrunch.com) The docs also describe voice input and cryptographic identities for users and agents. (github.com) The same documentation says Osaurus offers a Linux sandbox on macOS 26+ that can run shell, Python and Node workloads in an isolated virtual machine. TechCrunch reported that Osaurus uses a hardware-isolated virtual sandbox as part of its security model. ### What hardware does local use require? TechCrunch reported that running local models requires at least 64GB of RAM. The same report said Pae recommends about 128GB of RAM for larger models such as DeepSeek v4. (docs.osaurus.ai) Pulse24, citing TechCrunch, said Osaurus supports models including MiniMax M2.5, Gemma 4 and DeepSeek V4, and integrates more than 20 native macOS plugins. (docs.osaurus.ai) The Osaurus documentation separately says the platform includes “20+ native plugins,” including Mail, Calendar, Vision, Browser and Git. ### Where can users get it now? GitHub says users can install Osaurus with `brew install --cask osaurus` or download a DMG from Releases. (techcrunch.com) The repository page also links to documentation, models information, a Discord server and the project’s social accounts. The project documentation says onboarding is five steps and includes choosing a local, Apple Foundation or cloud model on first launch. (pulse24.ai) As of the latest repository snapshot, GitHub showed 5,200 stars and recent release-related updates in the main repository. (github.com)