Lume 0.2: Unattended macOS VMs for AI Agents and CI/CD

Lume 0.2 has arrived, introducing a suite of powerful features designed to automate the creation and management of macOS virtual machines on Apple Silicon. The open-source CLI tool, part of the Cua Computer Use Agent SDK, now enables fully unattended setup, persistent API-driven management, and direct integration with AI coding agents.

This release addresses a critical need for developers and researchers running AI agents in isolated macOS environments. By eliminating manual configuration steps and providing robust automation tools, Lume 0.2 streamlines workflows for everything from CI/CD pipelines to security research sandboxes.

The centerpiece of this release is Unattended Setup, a feature that takes a macOS IPSW file and produces a fully configured virtual machine without any keyboard input. This is achieved through a custom VNC and OCR system that automatically navigates the macOS Setup Assistant.

Alongside automation, Lume now runs as a background daemon with a REST API on port 7777. This allows scripts and continuous integration systems to manage VMs persistently, even after the original terminal session closes. The API supports commands to start, stop, and configure virtual machines programmatically.

lume create my-vm --os macos --ipsw latest --unattended tahoe

Users can define custom YAML configurations to tailor any macOS version to their specific requirements, ensuring consistent environments across different projects and teams.

For AI development, Lume 0.2 introduces a native MCP Server that integrates directly with Claude Desktop and other AI coding agents. This allows large language models to create, run, and execute commands inside macOS VMs through simple conversational prompts.

Addressing the perennial issue of limited disk space, the update adds support for multi-location storage. Developers can now move virtual machines to external SSDs or different volumes, cloning and managing images across various storage backends with simple commands.

• Unattended Setup: Automated macOS configuration via VNC/OCR
• HTTP API: Persistent VM management on port 7777
• MCP Server: Native Claude Desktop and AI agent integration
• Multi-location Storage: Support for external drives and SSDs
• Registry Support: Push and pull images from GHCR or GCS

These features collectively enable the creation of golden images that can be shared across teams via registries like GHCR or GCS, ensuring everyone works with identical, version-controlled environments.

Development teams are already leveraging Lume 0.2 for a variety of demanding tasks. The ability to run Claude Code in an isolated VM keeps the host system clean while allowing mistakes to be reset instantly by cloning from a known-good state.

For Apple platform app development, Lume provides a robust foundation for CI/CD pipelines. Automated UI testing can be performed across multiple macOS versions simultaneously, while security researchers utilize the tool to create disposable sandboxes for analyzing potentially malicious software.

lume run sandbox --shared-dir ~/my-project

The tool is built directly on Apple's native Virtualization Framework, requiring no emulation. This ensures high performance and compatibility with M1, M2, M3, and M4 chips, though it remains Apple Silicon exclusive.

Lume extends beyond local machines, with confirmed compatibility on EC2 Mac instances and Scaleway for cloud infrastructure needs. The development team is actively working on a managed cloud offering to provide on-demand macOS compute for teams that require scalable, remote resources.

The project is part of the broader Cua initiative, an open-source Computer Use Agent SDK. Development is ongoing, with the team actively soliciting feedback, bug reports, and feature ideas from the community to guide future iterations of the tool.

Installation is straightforward via a single command, and comprehensive documentation is available online. The entire project is licensed under the MIT License, encouraging widespread adoption and contribution.

Lume 0.2 represents a significant leap forward in automating macOS virtualization on Apple Silicon. By removing manual setup barriers and integrating deeply with AI agent workflows, it opens new possibilities for development, testing, and research.

Key advancements include the unattended setup process, the persistent HTTP API, and the MCP Server for AI integration. These features, combined with flexible storage and registry support, create a comprehensive platform for managing isolated macOS environments at scale.