CI Runnerscirunners.org
Runner profile · Runs in your cloud or on your hardware

Tart

Answer in brief

Virtualization toolset for macOS and Linux VMs on Apple silicon, with OCI-registry images and the Orchard orchestrator for clusters of Macs. Tart is a source-available VM toolset, now maintained by OpenAI, for teams that operate their own Apple silicon Macs and want image-based ephemeral macOS VMs for CI.

By the CI Runners editors · Updated

Overview

Tart is a command-line toolset for building, running and managing macOS and Linux virtual machines on Apple silicon, using Apple’s Virtualization.Framework. VM images are stored in OCI-compatible container registries, so a host clones an image much as it would pull a container. Orchard, its companion orchestrator, runs Tart VMs across a cluster of Macs through a controller, workers and a REST API.

Tart was created by Cirrus Labs, which announced on April 7, 2026 that it had agreed to join OpenAI. The repositories now sit under the openai GitHub organisation and, per tart.run, Tart has been maintained by OpenAI under FSL-1.1-ALv2 since June 5, 2026, with no licence fees. Documented CI integrations are a GitLab Runner custom executor and a Buildkite plugin; the GitHub Actions entry on tart.run points to the Cirrus Runners service.

Notable

  • April 7, 2026: Cirrus Labs announced an agreement to join OpenAI as part of its Agent Infrastructure team and said it had stopped charging licensing fees for Tart, Vetu and Orchard
  • June 5, 2026: Tart and Orchard were relicensed under FSL-1.1-ALv2; tart.run says Tart is maintained by OpenAI as of that date
  • February 11, 2023: Tart moved from AGPL-3.0 to Fair Source 100, a change tart.run now marks as historical
  • November 17, 2026: Tart 2.42.0 released on GitHub

Tart specifications

AttributeDetails
CI systemsGitLab CI, Buildkite, GitHub Actions
Operating systemsmacOS, Linux
Architecturesarm64
DeploymentRuns in your cloud or on your hardware · Self-hostable
Where it can runBare metal
IsolationJobs run in macOS or Linux virtual machines created with Apple’s Virtualization.Framework; the documented GitLab executor and Buildkite plugin run each job in an isolated ephemeral VM cloned from an image
Fresh environment per jobYes
HardwareRuns on Apple silicon Macs with macOS 13.0 (Ventura) or later; Orchard workers run on macOS and the Orchard controller can also run on Linux or Kubernetes.
Performance (vendor claim)The Tart README says Apple’s Virtualization.Framework gives near-native performance (project claim)
Runner sizesNot documented
CachingVM images are pushed to and pulled from OCI-compatible container registries. The GitLab executor example sets builds and cache directories inside the VM.
Docker layer cachingNot documented
GPUNot documented
ConcurrencyThe Orchard announcement says a single host allows a maximum of two virtual machines at a time; Orchard schedules VMs across a cluster of hosts
Job limitsNot documented
Static IPsNot documented
SSH into running jobsYes
ObservabilityOrchard controller and workers emit OpenTelemetry metrics.
SetupInstall with Homebrew (brew install openai/tools/tart) and clone a macOS image from a registry. Add the GitLab Tart Executor to a self-hosted GitLab Runner or the Tart plugin to a Buildkite pipeline; Orchard adds a controller and workers for several hosts.
Example labelimage: ghcr.io/cirruslabs/macos-tahoe-base:latest
PricingNo licence fees: Cirrus Labs said on April 7, 2026 that it had stopped charging licensing fees for Tart, Vetu and Orchard. You supply and pay for the Mac hardware.
Free tierFree to use under FSL-1.1-ALv2; the earlier paid tiers no longer apply, per tart.run
ComplianceNot documented
RegionsWherever your Apple silicon hosts run
Open sourceNo (FSL-1.1-ALv2)
Company headquartersSan Francisco, California, United States · More from the United States

Facts from Tart’s public documentation, verified November 22, 2026. Performance figures are vendor claims.

Strengths and trade-offs

Reasons to choose it

  • You already own or rent Apple silicon Macs and want to run CI jobs in disposable VMs on them
  • You want VM images versioned and distributed through an OCI container registry
  • You use GitLab CI or Buildkite, for which Tart documents an executor and a plugin
  • You need to schedule VMs across many Macs: Orchard provides a controller, workers and a REST API
  • You want to read the source: both Tart and Orchard are public on GitHub

Trade-offs

  • FSL-1.1-ALv2 is a source-available licence that excludes Competing Use; each version becomes Apache-2.0 on the second anniversary of its release
  • A single host allows a maximum of two virtual machines at a time, per the Orchard announcement
  • Only one Orchard controller instance is supported at the moment, per the Orchard docs
  • The GitHub Actions integration listed on tart.run is Cirrus Runners, which Cirrus Labs said on April 7, 2026 is no longer accepting new customers

Good fit for: Teams that run their own Apple silicon Macs and want scriptable, image-based macOS VMs for GitLab CI or Buildkite jobs.

Use cases Tart fits

  • ARM64 builds: Tart also runs arm64 Linux VMs (Ubuntu, Debian, Fedora images) on Apple silicon hosts.
  • macOS and iOS builds: Tart runs macOS VMs on Apple silicon from prebuilt images that include Xcode, for use in CI and other automation.
  • Your own cloud: Tart and Orchard run entirely on Mac hardware you operate.
  • Open source and self-hosted: The source is public and self-hosted, under the source-available FSL-1.1-ALv2 licence rather than an OSI-approved one.
  • GitLab CI runners: The GitLab Tart Executor runs GitLab jobs in ephemeral Tart VMs through GitLab Runner’s custom executor.

How Tart compares

RunnerCI systemsRuns onDeploymentIsolationPriceLargest runnerDocker layer cacheGPUOpen sourceCompany HQ
TartGitLab CI, Buildkite, GitHub ActionsmacOS, Linux, arm64Your cloud or hardwareApple Virtualization VMNot documentedSet by your infrastructureNot documentedNot documentedNoUnited States
OrkaGitHub Actions, GitLab CI, Buildkite, Jenkins, Azure Pipelines, TeamCitymacOS, arm64Your cloud or hardwaremacOS VMFrom $109/mo per hosted Mac mini (Orka by quote)Set by your infrastructureNot documentedNot documentedNoUnited States
GetMacGitHub Actions, GitLab CI, BuildkitemacOS, arm64Vendor-hostedmacOS VM$0.012/min (macOS M4, Developer plan)Not documentedNot documentedNot documentedNoLithuania
GitLab RunnerGitLab CILinux, macOS, Windows, x64, arm64Self-hostedDepends on executorFree software + your infrastructureSet by your infrastructureNot documentedYesYesNot documented
Actions Runner Controller (ARC)GitHub ActionsLinux, WindowsSelf-hostedKubernetes podFree software + your infrastructureSet by your infrastructureNot documentedNot documentedYesUnited States
Elastic CI Stack for AWSBuildkiteLinux, Windows, x64, arm64Self-hostedEC2 instanceFree software + your infrastructureSet by your infrastructureNot documentedNot documentedYesAustralia
GitHub-hosted runnersGitHub ActionsLinux, macOS, Windows, x64, arm64Platform-hostedFresh VM$0.006/min96 vCPU / 384 GBNot documentedYesNoUnited States

Head-to-head: Orka vs Tart · Cilicon vs Tart · Tart vs Tartelet

See all Tart alternatives →

Frequently asked questions

What is Tart?
Virtualization toolset for macOS and Linux VMs on Apple silicon, with OCI-registry images and the Orchard orchestrator for clusters of Macs. Tart is a source-available VM toolset, now maintained by OpenAI, for teams that operate their own Apple silicon Macs and want image-based ephemeral macOS VMs for CI.
Which CI systems does Tart work with?
Tart runs jobs for GitLab CI, Buildkite and GitHub Actions, on macOS, Linux and arm64.
How does Tart isolate jobs?
Tart documents its isolation as: Jobs run in macOS or Linux virtual machines created with Apple’s Virtualization.Framework; the documented GitLab executor and Buildkite plugin run each job in an isolated ephemeral VM cloned from an image. Each job gets a fresh environment that is discarded afterwards.
Is Tart open source?
No, Tart is not open source. It does run in your own cloud account or on your own hardware.
How much does Tart cost?
No licence fees: Cirrus Labs said on April 7, 2026 that it had stopped charging licensing fees for Tart, Vetu and Orchard. You supply and pay for the Mac hardware. Check the official pricing page for current rates.
How do you set up Tart?
Install with Homebrew (brew install openai/tools/tart) and clone a macOS image from a registry. Add the GitLab Tart Executor to a self-hosted GitLab Runner or the Tart plugin to a Buildkite pipeline; Orchard adds a controller and workers for several hosts. An example label is image: ghcr.io/cirruslabs/macos-tahoe-base:latest.
How fast is Tart?
The Tart README says Apple’s Virtualization.Framework gives near-native performance (project claim). Speed figures are the provider’s own and depend on the workload.
Does Tart have a free tier?
Free to use under FSL-1.1-ALv2; the earlier paid tiers no longer apply, per tart.run.
Does Tart support ARM64 runners?
Yes, Tart documents arm64 runners (macOS and Linux).
Does Tart offer macOS runners?
Yes, Tart documents macOS runners alongside Linux.
How many jobs can run at once on Tart?
The Orchard announcement says a single host allows a maximum of two virtual machines at a time; Orchard schedules VMs across a cluster of hosts.
Where do Tart runners run?
Wherever your Apple silicon hosts run.
Where is OpenAI based?
OpenAI, the organisation behind Tart, is headquartered in San Francisco, California, United States. That is where the company is based, not where jobs run; for that, see the regions listed for Tart.
What are the alternatives to Tart?

Sources

  1. cirruslabs.org/
  2. tart.run/
  3. tart.run/quick-start/
  4. tart.run/blog/2023/02/11/changing-tart-license/
  5. tart.run/blog/2023/04/25/announcing-orchard-orchestration-for-managing-macos-virtual-machines-at-scale/
  6. tart.run/integrations/gitlab-runner/
  7. tart.run/integrations/buildkite/
  8. tart.run/orchard/architecture-and-security/
  9. tart.run/orchard/deploying-controller/
  10. tart.run/orchard/managing-cluster/
  11. github.com/openai/tart
  12. github.com/openai/tart/blob/main/LICENSE
  13. github.com/openai/tart/releases
  14. github.com/openai/orchard
  15. openai.com/policies/privacy-policy/

Check before you rely on it. Products change quickly, and this page reflects public documentation as of November 22, 2026. Confirm anything that matters in the provider’s own documentation.

Cite this page

CI Runners. (2026). Tart. cirunners.org. Last reviewed November 22, 2026. https://cirunners.org/runners/tart/