Eaon: Free Mac app unifies 49 AI models, runs offline

Eaon is a single-window native macOS application designed to serve as a unified hub for AI model access across multiple providers and architectures. The app supports 49 hosted models out of the box: Claude, GPT-5, Gemini, Llama, Mistral, DeepSeek, Grok, Qwen, GLM, Kimi, and Gemma, among others. Users can switch between them via a single menu without managing separate logins; each conversation can even be split side-by-side to compare two models' answers on the same prompt.

Three pathways are available for connecting models. First, users can leverage Eaon's built-in hosted connection, paying only for usage with those providers. Second, they can bring their own API keys from OpenAI, Anthropic, or Google; these keys are stored in the macOS Keychain and never transmitted to a server. Third, they can run local open-source models entirely offline using Ollama, llama.cpp, or MLX—Eaon auto-detects these tools, allows users to download quantized models from Hugging Face or Ollama's registry, and handles warm-keeping and unloading of models in the background.

The interface emphasizes keyboard efficiency: a command palette (invoked with ⌘K) allows users to jump to chats, switch models, and adjust settings without reaching for the mouse. The app includes per-model token counts, live cost estimates, and a context-window meter so users can track spending and avoid surprises in long conversations. Chat history is stored as local files that can be exported or deleted at any time. The design follows macOS conventions—a monochrome native window that respects system theme, no ads, and no feeds.

Eaon launched as open source under a model where anyone can star the repository, open issues, or submit pull requests. The developer is a single indie contributor who has not yet obtained Apple's $99/year developer account needed for notarization. Consequently, on first launch, macOS flags the app as unsigned; users must either run a one-line Terminal command (`xattr -rd com.apple.quarantine /Applications/Eaon.app`) or go through System Settings to allow it. The developer is seeking $99 in annual donations via Ko-fi specifically to enable notarization, which would remove this friction for all future users. The app requires macOS 14 Sonoma or later; local models perform best on Apple silicon with sufficient memory for the chosen model. Support is available via email (support@eaon.dev) or Discord, with direct contact to the creator at sanscreates@eaon.dev.

Eaon emerges at a moment when many power users feel fragmented across multiple AI platforms. Instead of maintaining separate browser tabs for Claude (via Anthropic), ChatGPT, Google Gemini, and open-source alternatives, the app consolidates them into one keyboard-driven native interface. The developer's decision to build a native macOS application rather than another browser extension reflects a deliberate rejection of web-based limitations: a command palette (⌘K), system-theme integration, and the ability to run models completely offline are features that browsers cannot easily provide.

The offline-first design deserves particular attention for users handling sensitive work. By supporting Ollama, llama.cpp, and MLX, Eaon enables researchers, analysts, and others to keep conversations entirely on their machine—no API calls, no cloud logs. This is complemented by strict key management: API credentials live in macOS Keychain rather than plain text, and users can export or delete their full chat history anytime. The open-source approach (with a live GitHub download counter) reinforces transparency about adoption and invites community contribution.