2× faster CI runners.
With GPUs that don’t idle.

Performance target for the high-frequency runner fleet. Results vary by workload.

THE SWITCH
.github/workflows/ci.yml
jobs:
  test:
-   runs-on: ubuntu-latest
+   runs-on: purple-4vcpu-ubuntu-2404
    steps:
      - uses: actions/checkout@v4
      - run: npm test
2× fasterhigh-frequency CPU runners
GPU-nativeCUDA through VirtualGPU
Ephemeralone clean runner per job
Drop-inchange one workflow label
HOW IT WORKS

Pay for GPU time you use.
Not build time.

Purple runs setup, builds, and tests on fast CPU, then meters GPU acceleration only when your job calls for it.

01

GitHub sends the job

The App listens only for jobs carrying a purple-* label and records an idempotent queue event.

02

An idle node claims it

A cloud node leases the oldest compatible job and receives a GitHub JIT configuration valid for that runner only.

03

The runner disappears

The ephemeral runner executes one job, streams logs to GitHub, cleans its workspace, and deregisters automatically.

PRICING PREVIEW

Choose the GPU you need
for your Actions.

CPU time and GPU time are metered separately. Every job pays the CPU runner rate; the selected GPU rate is added only while accelerated work runs.

CPU TIME$0.01 / runner minute
+
GPU TIME$0.00–$0.10 / GPU minute
CPU only
$0.00GPU minute

Pay only the CPU runner rate.

RTX 4090
$0.01GPU minute

Fast CUDA for everyday workflows.

RTX 5090
$0.02GPU minute

Next-generation desktop GPU speed.

RTX Pro 6000
$0.03GPU minute

More memory for larger workloads.

H100
$0.05GPU minute

Datacenter acceleration for serious jobs.

Illustrative preview pricing. Final rates may change before general availability.

YOUR NEXT BUILD HAS SOMEWHERE TO GO

Put the purple machines to work.

Install in minutes. Pick the repositories. Change one label.

Install Purple on GitHub