StarSling Runners
Faster hardware, and agents that keep making it faster.
5th Gen AMD EPYC, from 2 to 64 vCPU, at 21% to 33% less per minute. Then agents read your runs and open pull requests that make them faster still.
Migrate in one prompt, review one diff
Your agent swaps runs-on: to StarSling labels across every workflow. Every changed line is a runs-on: line.
jobs:
build:
runs-on: ubuntu-latestjobs:
build:
runs-on: starsling-ubuntu-26.04Read https://docs.starsling.dev/start.md then set me up on StarSling.Faster hardware and AI agents
Hardware
5th Gen AMD EPYC
Every size runs Ubuntu 26.04 or 24.04 on 5th Gen AMD EPYC, so the same job runs about 1.7× faster before you change anything else about it.
Adaptive Compute
Sandbox Router sends every job to the fastest available sandbox, and Afterburner adds up to 25% more CPU and up to 50% more RAM to a running step at no extra cost. Up to 2.32x faster than GitHub-hosted runners.
Agents
Agent-generated PRs that optimize your CI
StarSling reads your real runs and opens pull requests that shard suites, fix cache configuration and cut critical path. The runners get faster after the swap, not just at it.
Agent-first CLI for your CI
sling is the agent-first CLI for StarSling Runners: why a job failed, where the time went and what CI costs, and the skill teaches your agent to use it.
Agent skills that optimize your CI
Open source, one install, and they run on your machine: each skill reads your workflows and hands your coding agent the evidence to fix them.
“That's why you just auto-accept the StarSling PR… they'd already proved it out for you.”
Daniel WorkuCo-founder & CTO, StarSling
Real PRs from StarSling agents
Pull Requests merged by customers. StarSling agents analyze and optimize your CI automatically.
Merged
1.8× faster ClickHouse store tests
Median of 5 runs per arm on the same commit: 530s → 298s. The PR estimates about 195 hours of CI time saved per 30 days.
perf(clickhouse): read the serial-cursor query results, clickhouse store tests 44% faster
View public PRMerged
Up to 5.4× faster
Mongo adapter Vitest duration: 24.3s → 4.5s in the PR’s benchmark, with 50 runs per arm.
perf(test-utils): stop re-deleting cleaned rows, up to 5.4x faster
View public PRMerged
≈18% faster test job
Mean test job on Node 22: 359s → 296s, and on Node 24: 342s → 280s, across 50 runs per arm (200 jobs total).
perf(test): use a faster default password hasher in getTestInstance
View public PRMerged
88% less checkout data
Full-history clone transfer: 1,572 MB → 189 MB. The PR estimates 445 runner-hours saved per month.
perf(ci): fetch full-history checkouts with --filter=blob:none
View public PRMerged
Publishing tokens scoped to publish jobs
Removed OIDC token access from the nonpublishing job; the three publishing jobs retain it.
ci(security): scope id-token: write to publish jobs only
View public PRMerged
≈3.4 minutes saved per run
Kitchen-sink E2E benchmark: 467s → about 260s for the slowest shard across five warm-cache runs.
ci: shard E2E kitchen-sink across 3 parallel jobs
View public PR
Six sizes, every one cheaper than its GitHub-hosted twin
Every label below runs Ubuntu 26.04 or 24.04 on 5th Gen AMD EPYC. Click one to copy it, then paste it after runs-on:. Prices are per minute, and GitHub's are its Linux x64 larger-runner list prices.
| Label | Size | StarSling | GitHub | You save |
|---|---|---|---|---|
| 2 vCPU / 8 GB | $0.004/min | $0.006/min | 33% | |
| 4 vCPU / 16 GBdefault | $0.008/min | $0.012/min | 33% | |
| 8 vCPU / 32 GB | $0.016/min | $0.022/min | 27% | |
| 16 vCPU / 64 GB | $0.032/min | $0.042/min | 24% | |
| 32 vCPU / 128 GB | $0.064/min | $0.082/min | 22% | |
| 64 vCPU / 256 GB | $0.128/min | $0.162/min | 21% |
| Label | Size | StarSling | GitHub | You save |
|---|---|---|---|---|
| 2 vCPU / 8 GB | $0.004/min | $0.006/min | 33% | |
| 4 vCPU / 16 GBdefault | $0.008/min | $0.012/min | 33% | |
| 8 vCPU / 32 GB | $0.016/min | $0.022/min | 27% | |
| 16 vCPU / 64 GB | $0.032/min | $0.042/min | 24% | |
| 32 vCPU / 128 GB | $0.064/min | $0.082/min | 22% | |
| 64 vCPU / 256 GB | $0.128/min | $0.162/min | 21% |
ubuntu-latest maps to starsling-ubuntu-26.04, ubuntu-26.04 maps to starsling-ubuntu-26.04, ubuntu-24.04 maps to starsling-ubuntu-24.04.
For the latest labels and pricing, see the StarSling docs.
Up to 94% off your GitHub Actions bill.
Faster runs, billed at up to 33% less per minute.
Give your coding agent the same playbook
The agents open pull requests on their own. These are the same skills as commands, free and open source, for the times you want to ask.
/ci-speedupci-speedup samples your own Actions runs, finds the check that gates your pull requests, and hands your coding agent the evidence to shorten it.See how ci-speedup works/ci-scoreci-score reads a local checkout, grades it against the best-practice rubric we publish, and ranks a fix for every gap it finds.See how ci-score works/ci-secureci-secure scans your workflows for the ten critical attack vectors, explains each as a plain-English attacker scenario, and fixes the ones you pick.See how ci-secure works/slingsling is the agent-first CLI for StarSling Runners: why a job failed, where the time went and what CI costs, and the skill teaches your agent to use it.See how sling works
Runner and billing FAQ
How does StarSling work with GitHub Actions?
StarSling provides compute for your GitHub Actions jobs. Keep your workflows, triggers, actions and GitHub orchestration, and choose a StarSling runs-on label.
Which CPU labels can I use?
Ubuntu 26.04 and Ubuntu 24.04 run on x64 5th Gen AMD EPYC machines. Supported labels: starsling-ubuntu-26.04-2, starsling-ubuntu-26.04, starsling-ubuntu-26.04-8, starsling-ubuntu-26.04-16, starsling-ubuntu-26.04-32, starsling-ubuntu-26.04-64, starsling-ubuntu-24.04-2, starsling-ubuntu-24.04, starsling-ubuntu-24.04-8, starsling-ubuntu-24.04-16, starsling-ubuntu-24.04-32, starsling-ubuntu-24.04-64.
Which GPU runners can I use?
Dedicated NVIDIA GPU runners are available by request for your organization: NVIDIA RTX PRO 6000: starsling-ubuntu-26.04-gpu; NVIDIA RTX 5090: starsling-ubuntu-26.04-gpu/gpus=rtx-5090:1; NVIDIA RTX 4090: starsling-ubuntu-26.04-gpu/gpus=rtx-4090:1; NVIDIA RTX PRO 6000: starsling-ubuntu-24.04-gpu; NVIDIA RTX 5090: starsling-ubuntu-24.04-gpu/gpus=rtx-5090:1; NVIDIA RTX 4090: starsling-ubuntu-24.04-gpu/gpus=rtx-4090:1. Each GPU rate includes a 4 vCPU, 16 GB RAM host and 100 GB disk.
How is runner time billed?
New paid subscriptions are billed in whole seconds from job start to finish. Partial seconds round up, with no one-minute minimum. Existing paid subscriptions remain billed in whole minutes with a one-minute minimum until migrated. Queue time is never billed. Concurrency is unlimited. New accounts get 2,000 free minutes in their first month. Failed jobs are billed for the time they run: whole seconds on new paid subscriptions, with partial seconds rounded up; whole minutes on existing paid subscriptions until migrated.
How are jobs isolated?
Every job runs on a dedicated, single-use virtual machine. Customer isolation uses the virtual-machine boundary. GitHub injects secrets into the runner; workflow logs and published Disks follow their documented retention policies. Read the security overview for isolation, data handling, GitHub App permissions, and compliance status.
How do optimization agents activate?
Optimization agents analyze workflows, logs and telemetry and submit pull requests for your engineers to review and merge. Paid-plan customers can activate optimization PRs for their organization; new accounts start with optimization PRs disabled.
How are Review Runners billed?
Review Runners use starsling-review, with 4 vCPU and 16 GB memory. Review Runner time has a normalized rate of $0.004 per minute. New paid subscriptions are billed in whole seconds from job start to finish. Partial seconds round up, with no one-minute minimum. Existing paid subscriptions remain billed in whole minutes with a one-minute minimum until migrated. Queue time is never billed. Concurrency is unlimited. Model tokens are billed by your model provider on your own account.
References: Billing and rates, runner labels, data handling.
Same YAML, different silicon
Start faster. Then keep getting faster.
Sign up with GitHub and install the StarSling GitHub App, point a workflow at a StarSling label, and compare the two runs side by side. The agents start reading your runs from the first one. New accounts get 2,000 free minutes in their first month. No credit card required.
