OpenAI Acquires Cirrus Labs: Tart Free, CI Shuts June 1

Cirrus Labs — a bootstrapped, 9-year-old engineering team behind macOS virtualization and CI/CD tooling that never took external funding — announced on April 7, 2026 that it's joining OpenAI's Agent Infrastructure team, the division building the environment layer that AI automation agents need to operate. The news hit #1 on Hacker News with 113 points and 45 comments within hours. But buried in the announcement are two facts that affect thousands of developers right now: Cirrus CI is shutting down on June 1, 2026, and the remaining tools are being relicensed to become more permissive — not less.

What Cirrus Labs Built Over 9 Years — Without a Single Investor

Fedor Korotkov — previously an engineer at Airbnb and Twitter — founded Cirrus Labs in 2017 with a simple goal: build better CI/CD (continuous integration and deployment — the automated systems that test and ship code every time a developer makes a change) tooling for the cloud era. Unusually for a developer tools company, Cirrus Labs never raised outside capital, which gave it the freedom to build at its own pace and stay focused on quality.

Over nine years, the team shipped four major products:

• Cirrus CI — launched 2018, one of the first CI/CD platforms (automated testing pipelines) to support Linux, Windows, and macOS in a single system. Offered free minutes for open-source projects. Shutting down June 1, 2026.
• Tart — launched 2022, a virtualization tool (software that creates isolated "virtual computers" running macOS inside your Mac) for Apple Silicon (M1, M2, M3, M4 processors). Now has 5,300+ GitHub stars and is used in production by Atlassian, Figma, Expo, Mullvad, Krisp, and Transloadit.
• Orchard — an orchestration layer (a management system for automatically running large fleets of VMs simultaneously) built on top of Tart. Enables teams to spin up hundreds of macOS environments on demand.
• Vetu — a complementary virtualization tool in the Cirrus Labs ecosystem.

Tart's performance advantage comes from using Apple's own Virtualization.Framework — a low-level API (programming interface) built into macOS that runs VMs at near-native speeds, rather than the slower software emulation approach older tools used. Running a full macOS Ventura VM inside Tart on an M2 Mac Mini performs almost identically to running on bare metal hardware.

The project is written primarily in Swift (90.6%) — Apple's native language — which is rare in the open-source infrastructure world and represents years of specialized Apple platform expertise that is genuinely hard to hire.

Why OpenAI Acquired Cirrus Labs — AI Agent Infrastructure at Scale

The acquisition announcement included a direct quote from Korotkov explaining the fit: "This allows us to extend the mission we started with Cirrus Labs: building new kinds of tooling and environments that make engineers more effective — for both human engineers and agentic engineers."

This is more specific than it sounds. OpenAI's AI agents — systems that can autonomously browse the web, write code, run tests, and operate software — need isolated environments (sandboxed spaces where the agent's actions can't affect other systems) to work safely. Running an agent that controls a real browser or desktop requires a fresh, isolated desktop environment for every task.

Tart and Orchard solve exactly this problem for macOS. Key reasons OpenAI wants this team on its Agent Infrastructure team:

• macOS at scale — OpenAI's computer use features (where AI models take control of real software interfaces — clicking buttons, typing text, reading screens) need real, isolated macOS desktops. Tart lets you spin up a fresh macOS VM in seconds; Orchard manages hundreds of them simultaneously.
• Apple Silicon efficiency — Apple's M-series chips deliver roughly 2–3× better performance-per-watt than x86 server hardware for certain inference workloads (the process of running a trained AI model to produce outputs). A rack of Mac Mini M4 Pro machines running Tart VMs is significantly more energy-efficient for macOS agent tasks than equivalent x86 setups.
• Near-native VM performance — Cirrus Runners, the commercial runner product built on Tart, delivered "2–3× better performance for a fraction of the price" compared to GitHub Actions' standard hosted macOS runners. This benchmark matters when you're running thousands of agent test cycles per day.
• Rare platform expertise — The Cirrus Labs team spent years mastering Apple's Virtualization.Framework, Swift-based systems programming, and macOS VM lifecycle management. This expertise is not easily hired or replicated.

The Agent Infrastructure Layer

OpenAI's "Agent Infrastructure team" is not a well-publicized division, but its function is clear: it builds the plumbing that AI agents run inside. Compute environments, sandboxing, VM lifecycle management, snapshot and restore capabilities (saving and loading exact VM states to resume tasks instantly) — these are the unsexy but critical systems that determine whether autonomous AI agents are usable in production. Cirrus Labs' entire portfolio maps directly onto this need.

Cirrus CI Shuts Down June 1 — You Have 7 Weeks to Migrate

If you're running CI/CD pipelines on Cirrus CI today, this is the section that matters most. The service is shutting down on June 1, 2026 — that's less than 7 weeks from now. Cirrus Labs has also stopped accepting new customers for Cirrus Runners.

Your migration options, ranked by effort:

• GitHub Actions macOS runners — the path-of-least-resistance for most teams. Syntax is different but the migration is well-documented. macOS runners on GitHub Actions are more expensive per minute than Cirrus was, but you're already paying for GitHub.
• Self-hosted Tart + Orchard — now entirely free (licensing fees removed as part of the OpenAI transition). If you have Apple Silicon hardware, this is the most cost-effective long-term option and gives you the same near-native performance Cirrus Runners offered. The Orchard documentation covers multi-machine setups.
• Buildkite with macOS agents — a commercial CI platform with strong macOS support and Tart integration available.
• Cirrus CI config compatibility — Cirrus CI uses a .cirrus.yml format. Most tasks translate directly to GitHub Actions syntax; the main difference is how environment variables and matrix builds (running the same tests across multiple configurations simultaneously) are declared.

Here's the minimal Tart setup to get a macOS VM running locally as a self-hosted replacement:

# Install Tart via Homebrew
brew install cirruslabs/cli/tart

# Pull a macOS base image (Ventura)
tart clone ghcr.io/cirruslabs/macos-ventura-base:latest ventura-base

# Start the VM
tart run ventura-base

# List running VMs
tart list

The full Orchard setup for multi-machine orchestration (running many VMs across several Mac Minis automatically) takes more configuration but is now freely available in the Orchard GitHub repository.

The Surprise: Tools Get More Permissive, Not Less

Every acquisition of a developer tools company triggers the same fear: the acquirer locks down the open-source software, changes the license, and starts charging for what used to be free. HashiCorp did it with Terraform in 2023 (triggering the OpenTofu fork). Docker did it in 2022. Redis did it in 2024.

Cirrus Labs is going in the opposite direction. As part of joining OpenAI:

• Tart and Orchard are being relicensed to more permissive terms
• Licensing fees are being removed — previously, commercial use of certain Cirrus Labs tools required a paid license
• The tools remain open-source and actively maintained (at least for now — OpenAI inherits the repository ownership)

This is the mirror image of the usual acquisition pattern. OpenAI appears to be making these tools freely available as ecosystem-building — the same logic that led Google to open-source TensorFlow or Meta to open-source PyTorch. If the surrounding ecosystem adopts your infrastructure, it strengthens your own platform. More developers using Tart means more macOS VM tooling built around OpenAI's emerging agent infrastructure standards.

The practical implication for developers: you can now use Tart and Orchard commercially without paying licensing fees. If you were avoiding Cirrus Labs tools due to cost, that barrier is gone. Check our AI automation guides for workflows that integrate macOS VM environments into your development pipeline.

Related Content — Get Started | Guides | More News