# Runners — High-Performance x64 and macOS GitHub Actions Runners (https://luxor.ai/products/runners)

> For the complete documentation index, see [llms.txt](https://luxor.ai/llms.txt)

# Cut your CI bill and build times

Runners for GitHub Actions. With one line of YAML. Bare-metal Linux and Apple Silicon macOS runners for GitHub Actions, billed by the second.

[Start for free](https://app.luxor.ai/auth/registration) · Onboard via Agent (copies a prompt that points your coding agent at [https://luxor.ai/skill.md](https://luxor.ai/skill.md))

- **3×** faster than GitHub
- **50%** of GitHub's price
- **$0.002** per vCPU / minute

Trusted by hundreds of engineering teams. Infrastructure, data and AI companies build on us.

## Half GitHub's per-minute price, up to 3× faster.

Benchmark comparisons of the same workflows on GitHub-hosted runners vs Luxor Agent Cloud.

x64 (Linux):

| Workflow                           | Luxor Runners time | GitHub time | Luxor Runners cost | GitHub cost | Faster | Cheaper |
| ---------------------------------- | ------------------ | ----------- | ------------------ | ----------- | ------ | ------- |
| PostHog/posthog docker build       | 4m 33s             | 14m 30s     | $0.0728            | $0.33       | 3.2×   | 78%     |
| nodejs/node build + test           | 37m 15s            | 1h 50m 41s  | $0.596             | $2.442      | 3.0×   | 76%     |
| torvalds/linux kernel build        | 2m 56s             | 7m 1s       | $0.0469            | $0.176      | 2.4×   | 73%     |
| apache/kafka build                 | 3m 48s             | 8m 34s      | $0.0608            | $0.198      | 2.3×   | 69%     |
| spring-projects/spring-boot build  | 47m 1s             | 1h 22m 58s  | $0.752             | $1.826      | 1.8×   | 59%     |
| kubernetes/kubernetes build + test | 11m 43s            | 19m 17s     | $0.375             | $0.84       | 1.6×   | 55%     |

macOS (Apple Silicon):

| Workflow         | Luxor Runners time | GitHub time | Faster |
| ---------------- | ------------------ | ----------- | ------ |
| iina/iina        | 2m 52s             | 8m 35s      | 3.0×   |
| nodejs/node      | 29m 54s            | 55m 10s     | 1.8×   |
| microsoft/vscode | 29m 54s            | 55m 10s     | 1.8×   |
| hashicorp/nomad  | 4m 50s             | 6m 16s      | 1.3×   |

And that's before the mirrors, which sit next to the runners and keep repeat installs off the public internet. [Learn about actions cache](https://luxor.ai/docs/runners/caching.md)

## One YAML line. That's it.

Larger x64 configurations are available on request. macOS runners run macOS 26 Tahoe on Apple Silicon.

Replace the `runs-on` value in your workflow with a Luxor Agent Cloud label:

```diff
 jobs:
   build:
-    runs-on: ubuntu-latest
+    runs-on: luxor-ubuntu-small-2c-4g
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-node@v4
```

Linux x64 · AMD Ryzen:

| `runs-on` label                   | vCPU | RAM   | Disk   | Price      |
| --------------------------------- | ---- | ----- | ------ | ---------- |
| `luxor-ubuntu-small-2c-4g`        | 2    | 4 GB  | 86 GB  | $0.004/min |
| `luxor-ubuntu-medium-4c-8g`       | 4    | 8 GB  | 150 GB | $0.008/min |
| `luxor-ubuntu-large-8c-16g`       | 8    | 16 GB | 200 GB | $0.016/min |
| `luxor-ubuntu-large-plus-16c-32g` | 16   | 32 GB | 300 GB | $0.032/min |

macOS · Apple Silicon M4 Pro:

| `runs-on` label         | vCPU | RAM   | Disk   | Price      |
| ----------------------- | ---- | ----- | ------ | ---------- |
| `luxor-macos-26-mini`   | 2    | 4 GB  | 140 GB | $0.040/min |
| `luxor-macos-26-small`  | 4    | 8 GB  | 140 GB | $0.080/min |
| `luxor-macos-26-medium` | 6    | 16 GB | 140 GB | $0.120/min |
| `luxor-macos-26-large`  | 8    | 32 GB | 140 GB | $0.160/min |
| `luxor-macos-26-xlarge` | 14   | 38 GB | 140 GB | $0.280/min |

[Learn what Runner size to use](https://luxor.ai/docs/runners/which-runner-should-i-use.md)

## Our own hardware, and the storage to match.

- **Run on bare metal** — 5th Gen AMD Ryzen for x64, Apple Silicon M4 Pro for macOS. Two vCPUs is one physical core, every time. [Bare metal docs](https://luxor.ai/docs/runners/sizes.md)
- **Start every job warm** — Dependency and compiler caches live on a disk that persists between jobs. Nothing to pack or upload. [Sticky disks docs](https://luxor.ai/docs/runners/sticky-disks.md)
- **Run jobs inside your tailnet** — Jobs on a dedicated Linux runner label join your Tailscale network before the first step runs, with no Tailscale action or secret in your workflow. You choose which repositories get access, and pull requests from forks stay out by default. [Read the Tailscale guide](https://luxor.ai/docs/runners/tailscale.md)

## Queue time is never billed. Pay for what runs.

Compute is billed by the second. Rates are shown per minute to compare with GitHub. [View pricing](https://luxor.ai/pricing)

| Rate                 | Tag       | Price                    |
| -------------------- | --------- | ------------------------ |
| x64 Runners          | AMD Ryzen | $0.002 per vCPU / minute |
| macOS Runners        | M4 Pro    | $0.020 per vCPU / minute |
| Cache & Sticky Disks | Storage   | $0.20 per GiB / month    |

Custom images and nodes are available with quote-based pricing. [Request a quote](mailto:hello@luxor.ai)

## Enterprise-grade security.

Every job runs in a single-use VM, destroyed when it ends.

- **Single-use VMs** — Destroyed at job end. Nothing carries over.
- **We don't store your secrets** — Your secrets stay in GitHub.
- **Isolation at the hypervisor** — Your own VM on bare metal, never shared.
- **Fixed egress range** — One fixed CIDR. No surprise rotation.

## Don't believe us. Believe them.

Customer quotes from founders, CTOs, and engineering managers on faster builds, lower CI costs, and easy migration. One example: "Been using Luxor Agent Cloud for a month now and it's incredible. 2x faster than github and cheaper." — Daniel Donaldson, SWE Manager.

## Questions? Answered.

Have a question we haven't answered? Reach out to hello@luxor.ai.

### What are Agent Cloud runners?

Agent Cloud runners are self-hosted-style GitHub Actions runners managed by Luxor Agent Cloud on bare-metal Linux x64 and macOS infrastructure, designed to be a drop-in alternative to GitHub-hosted runners.

You point your workflow at an Agent Cloud runner label by changing the `runs-on` value in your YAML, and your existing GitHub Actions jobs run unmodified on faster, cheaper hardware. There is no agent to install on your own servers and no infrastructure for your team to operate.

### How do I switch a workflow from GitHub-hosted runners to Luxor Agent Cloud?

Change the runs-on value in your workflow YAML to an Agent Cloud runner label (for example replace ubuntu-latest with luxor-ubuntu-small-2c-4g) and your next workflow run executes on Luxor Agent Cloud.

No other changes are required: Actions, secrets, caching, matrix strategies, reusable workflows, and container jobs all continue to work. The [Runners quickstart](https://luxor.ai/docs/runners/quickstart) walks through the migration wizard, which can also open the YAML pull request for you.

### Are Agent Cloud runners isolated and secure?

Yes. Every job runs in a single-tenant ephemeral virtual machine with a fresh kernel, fresh disk, and isolated network namespace, and the VM is destroyed at the end of the workflow.

Jobs cannot inspect or interfere with one another, and job-time secrets never leave the runner VM boundary. Luxor Agent Cloud is built on the same security controls and operational policies as Luxor's platform, which has completed a SOC 2 Type 2 examination, and a Luxor Agent Cloud SOC 2 report is in progress. See the [Security page](https://luxor.ai/company/security) for the full security program, encryption details, incident-response process, and report request flow.

### How much do Agent Cloud runners cost compared to GitHub-hosted runners?

Agent Cloud runners cost half GitHub's per-minute price and run the same workflows up to 3x faster than GitHub-hosted runners, with compute billed by the second.

Pricing is published on the [Pricing page](https://luxor.ai/pricing) and new accounts get $100 in free credits. Committed-use discounts are available for teams with steady or high-volume workloads.

### Which operating systems and architectures do Agent Cloud runners support?

Luxor Agent Cloud supports Linux on x64 and macOS on Apple Silicon (M-series) hardware, covering CI workloads for both general-purpose backend/web pipelines and teams that build, test, or sign Apple platform code.

Linux runners cover the same Ubuntu image surface as standard GitHub Actions, so most workflows that target `ubuntu-latest` migrate without source changes. For macOS specifics (Xcode toolchains, keychain handling, and signing) see the [macOS runners docs](https://luxor.ai/docs/runners/macos-runners).