Summary

This project provides an agentic AI framework for macOS, enabling deep integration with local and cloud LLMs to automate tasks, control applications, and manage codebases. It targets technically savvy users seeking a powerful, subscription-free alternative to existing AI development tools, offering comprehensive Mac control and AI-driven workflows.

How It Works

Agent! is a native macOS application built in Swift/SwiftUI that orchestrates 17 LLM providers, including on-device Apple Intelligence. It leverages macOS's Accessibility API (via AXorcist) and AppleScript for granular app control, dynamically discovers system and user-installed applications, and features a robust tool execution layer. Key architectural choices include XPC for privileged operations, prompt caching for efficiency, and a configurable fallback chain for LLM provider reliability. On-device Apple Intelligence handles initial triage, summarization, and UI automation, minimizing cloud LLM token usage.

Quick Start & Requirements

• Installation: Download the latest release and drag the application to /Applications.
• Requirements: macOS 26.4+ and Apple Silicon are mandatory. Building from source requires Xcode Command Line Tools or a full Xcode installation with an Apple Developer account for proper signing.
• Setup: Configure LLM provider API keys or select local providers in the app's Settings.
• Links: Latest Release

Highlighted Details

• Supports 17 LLM providers, including Claude, GPT, Gemini, Ollama, and on-device Apple Intelligence.
• Full macOS application control via the Accessibility API (AXorcist) and AppleScript/JXA.
• Native Xcode integration for building, debugging, code review, and version bumping.
• On-device Apple Intelligence for local task summarization, triage, and UI automation.
• Time-machine-style file rollback for all file edits, with a 1-week TTL.
• Remote control via iMessage and voice commands ("Agent!").
• Privacy-centric design with local data processing and no telemetry.

Maintenance & Community

The project is actively developed, with a call for community support via GitHub Stars and Forks. No specific community channels (e.g., Discord, Slack) or a public roadmap are detailed in the README.

Licensing & Compatibility

• License: MIT License, permitting free use, modification, and distribution.
• Compatibility: Strictly limited to macOS 26.4+ on Apple Silicon hardware. Commercial use is generally compatible under the MIT license terms.

Limitations & Caveats

Exclusive to macOS 26.4+ and Apple Silicon. Running large local LLMs necessitates high-end hardware (e.g., 64GB+ unified memory for 30B+ models). iMessage integration requires granting Full Disk Access. Building without an Apple Developer account results in ad-hoc signing, preventing Launch Agent/Daemon helper registration, though core LLM functionality remains unaffected.