Hedos Launches Native macOS App for Running Local AI Models with Full Privacy
Key Takeaways
- ▸Hedos is a native macOS app that runs all types of AI models (text, image, speech) locally on Apple Silicon with zero cloud dependency
- ▸The app automatically discovers existing models from Ollama, Hugging Face, and manual installations, then routes each through its optimal runtime engine
- ▸Complete privacy by design: no data, telemetry, or accounts required; all processing happens offline on the user's machine
Summary
Hedos, an open-source project, has released a native macOS application that enables users to discover and run large language models, image generation, and speech models directly on their machines with complete privacy. Built as a native Swift app optimized for Apple Silicon using MLX, the application supports models from multiple sources including Ollama, Hugging Face, and locally stored models, eliminating the need for cloud connectivity or external servers.
The v0.1.2 early release positions Hedos as a privacy-focused alternative to cloud-based AI services. The app automatically detects all models already installed on a user's Mac and intelligently routes each model through its optimal runtime engine—whether natively on Apple Silicon, through a managed runtime, or with a specific recipe. Users can download the application directly or install it via Homebrew, with both methods providing the native app and hedos command-line tool.
With its MIT license, transparent tier system showing how each model runs, and commitment to keeping all prompts, outputs, and models on the user's hardware, Hedos addresses growing concerns about data privacy in AI. The application requires no accounts, telemetry, or internet connection after setup, making it a fully self-contained solution for local AI inference.
- MIT open-source release with native Swift implementation on MLX foundation—not a web wrapper
Editorial Opinion
Hedos represents a meaningful shift in making AI accessible and private at the local level. By building a native, single interface for discovering and running all types of models on consumer hardware, it removes the friction between users and local inference while sidestepping the privacy and cost concerns of cloud AI. The emphasis on transparency—showing exactly how each model runs and what tier it occupies—is a refreshing contrast to black-box cloud services. For developers and users seeking full ownership of their AI tools, this open-source approach is a compelling alternative to vendor lock-in.
