Osaurus is an open-source, native macOS harness for AI agents built in Swift. Its pitch is simple: own your AI. It runs agents on your own Mac — with any model, persistent memory and autonomous execution — fully offline and under your control, rather than inside someone else’s cloud. The project has earned about 8,100 stars on GitHub and is under very active development, with new commits landing in October 2026.
Osaurus wraps local and remote models in a single native app designed for Apple Silicon. Local inference runs through MLX and Apple’s stack — including Apple Intelligence foundation models and the Neural Engine — while API models from providers such as Anthropic and OpenAI can be plugged in when you want them. Agents get persistent memory so work carries across sessions, tools through MCP (Model Context Protocol), and a cryptographic identity layer so agent actions can be attributed and trusted. Because the harness is a real Swift application rather than an Electron shell, it feels fast and at home on macOS, and everything — models, memory, files — stays on your machine unless you choose otherwise. It is aimed at people who want a personal, always-available agent runtime instead of renting one.
Osaurus is free and open source under the MIT license. If you connect paid API models, you pay those providers directly; local models cost nothing to run.
Osaurus is for Mac users and developers who want a private, self-owned AI agent that lives on their own machine instead of in a vendor’s cloud.