mac-mini-agent  by disler

macOS automation framework for AI agents

Created 1 month ago
256 stars

Top 98.5% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

Mac Mini Agent empowers AI agents with autonomous macOS device operation, bridging the gap between code generation and feature deployment. It enables agents to interact with GUIs, terminals, and applications, enhancing engineer productivity and agent autonomy. Aimed at mid-to-senior engineers, it facilitates complex, autonomous task execution across devices.

How It Works

Four CLIs form the core: Steer (Swift) provides macOS GUI automation via OCR and input simulation, crucially supporting Electron apps; Drive (Python) offers programmatic tmux control with a sentinel pattern for deterministic command completion; Listen (Python FastAPI) acts as a job server for agent task dispatch; and Direct (Python) is the client CLI. This integrated approach allows agents full control over the macOS environment.

Quick Start & Requirements

Automated setup scripts (/install-agent-sandbox, /install-engineer-devbox) manage dependencies (Swift, tmux, Python) across two machines: an agent sandbox (Mac Mini) and an engineer devbox. The agent sandbox requires manual granting of Accessibility, Screen Recording, Full Disk Access, and Remote Login permissions for Terminal via System Settings.

Highlighted Details

  • OCR for Electron Apps: Enables interaction with inaccessible UI elements in apps like VS Code, Slack, Notion.
  • Sentinel Pattern: Ensures reliable, deterministic command completion detection in tmux.
  • Cross-App Pipelines: Facilitates complex, multi-application workflows autonomously.
  • Agent-Agnostic: Compatible with any shell-invoking agent (e.g., Claude Code, Gemini CLI).

Maintenance & Community

The README provides no specific details on maintainers, community channels, or project roadmap.

Licensing & Compatibility

The project's license is not explicitly stated, posing a potential adoption blocker, especially for commercial use. The system is macOS-exclusive and mandates a two-machine setup.

Limitations & Caveats

This tool is macOS-specific and appears to be in early development (v0.1.0-v0.2.0). Setup requires manual configuration of sensitive macOS permissions on the agent execution machine. A mandatory two-machine architecture (agent sandbox + engineer devbox) is necessary. The absence of a clear license requires clarification.

Health Check
Last Commit

1 month ago

Responsiveness

Inactive

Pull Requests (30d)
1
Issues (30d)
2
Star History
32 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.