Osaurus launches native macOS AI agent harness

This is a timely and much-needed alternative to resource-heavy Electron-based and Python-dependent AI agent frameworks, proving that local agentic workflows can run smoothly and securely on consumer hardware when built natively. Building entirely in Swift with an MLX runtime bypasses Python overhead to keep memory consumption low, while keeping files, context, and memory strictly local appeals to users with strict privacy requirements. Furthermore, a built-in approval gate prevents agents from performing destructive actions without explicit consent, and model agnosticism allows seamless switching between local and cloud models while preserving memory.