macOS26/Agent

What it solves

Agent! is a native macOS application that transforms your desktop into an autonomous agentic AI environment. It solves the problem of AI assistants that can only talk about tasks; instead, it can execute actions directly on your Mac, such as writing and building Xcode projects, controlling other applications via the Accessibility API, and automating web browsers.

How it works

The project implements an autonomous task loop where the AI reasons, executes tools, and self-corrects based on observed results. It integrates with 18 different LLM providers (including local options like Ollama and on-device Apple Intelligence) and uses a native Swift-based ecosystem to minimize RAM usage and footprint. It employs a dedicated macOS Launch Daemon for privileged root-level command execution and uses the Accessibility API (via AXorcist) to drive any Mac app programmatically.

Who it’s for

Developers, power users, and macOS enthusiasts who want a fully autonomous AI agent that can manage their codebase, automate repetitive desktop tasks, and integrate with various AI models without relying on third-party subscriptions or telemetry.

Highlights

• Agentic Coding: Full environment for reading codebases, editing files with surgical diffs, and building Xcode projects with Time-Machine-style rollbacks.
• Desktop Automation: Ability to control any Mac app through the Accessibility API and automate Safari via JavaScript/AppleScript or Playwright.
• Multi-Provider Support: Compatibility with 18 LLM providers, including cloud APIs and fully local/on-device models.
• Remote & Voice Control: Support for a "Agent!" hotword for voice commands and remote control via iMessage.
• Self-Verifying Autonomy: Tasks require evidence-verified success criteria (goal_state) before they can be declared complete.
• MCP Server Support: Extensible via the Model Context Protocol to add additional tools.