The modern way to run GitHub Actions
Faster is good, cheaper is great, but that’s only half of the battle.
RunsOn is the only platform to come with fully integrated solutions for fast, unlimited caches, including docker layers. And no limit on concurrency!
| Â |
|---|
| Speed |
| Cost |
| Cache size |
| Cache speed |
| Concurrency |
| Docker cache |
| Custom runner images |
| x64 / ARM / MacOS |
| Secure |
| Static IPs |
| Account type |
| RunsOn |
|---|
| 400MB/s, local S3 backend cache |
 Unlimited concurrency |
| x64 + ARM64 |
| Personal + Organization |
| GitHub |
|---|
| Slow, even for larger runners |
| $$$ Pricey |
| 10GB cache per repo |
| 100MB/s cache bandwidth |
| 20 (Free plan), 60 (Team plan) |
 No native docker cache |
 No custom image |
| x64 + MacOS + Windows |
 Secure |
 Static IPs for Larger runners |
| Personal + Organization |
| Third-party SaaS |
|---|
| Up to 2x faster |
| 2x cheaper |
| 10-25GB cache per repo |
| 100-400MB/s cache bandwidth |
| 64vCPUs max, or you pay more |
 No native docker cache |
 No custom image |
| x64 + ARM + ~MacOS |
 Third-party, no SOC2 certification |
 No static IP |
| Organization only |
The RunsOn server is deployed in your AWS account and runs on a $1.5/month AWS AppRunner container. It receives workflow events from GitHub Action, and launches ephemeral EC2 instances to fulfill the workflow job requirements.
No dangling EC2 instances, unused capacity, docker-in-docker, or kubernetes cluster to manage (see architecture).
It can be deployed in 4 different regions, and takes advantage of multiple availability zones to make sure it will always find a spot instance, with the least probability of being interrupted. In case no spot instance is available, it automatically switches to on-demand pricing to ensure that the job is scheduled (still 5x cheaper than GitHub).
Note: this is the pricing with the most recent EC2 family types (m7a*,m7i*). You can get even more savings by switching to
previous-generation instance types.
| Runner | RunsOn | GitHub | RunsOn vs GitHub |
|---|---|---|---|
| 1cpu-linux-x64 | $0.0010 | - | - |
| 2cpu-linux-x64 | $0.0011 | $0.0080 | 7.2x cheaper |
| 4cpu-linux-x64 | $0.0018 | $0.0160 | 9.0x cheaper |
| 8cpu-linux-x64 | $0.0026 | $0.0320 | 12.4x cheaper |
| 16cpu-linux-x64 | $0.0055 | $0.0640 | 11.5x cheaper |
| 32cpu-linux-x64 | $0.0091 | $0.1280 | 14.1x cheaper |
| 48cpu-linux-x64 | $0.0137 | - | - |
| 64cpu-linux-x64 | $0.0155 | $0.2560 | 16.5x cheaper |
Note: this is the pricing with the most recent EC2 family types (m7g*). You can get even more savings by switching to previous-generation
instance types.
| Runner | Pricing |
|---|---|
| 1cpu-linux-arm64 | $0.0009 |
| 2cpu-linux-arm64 | $0.0012 |
| 4cpu-linux-arm64 | $0.0016 |
| 8cpu-linux-arm64 | $0.0024 |
| 16cpu-linux-arm64 | $0.0034 |
| 32cpu-linux-arm64 | $0.0062 |
| 48cpu-linux-arm64 | $0.0083 |
| 64cpu-linux-arm64 | $0.0120 |
RunsOn helped reduce our monthly CI costs from $1100 (third-party SaaS) to $400. Also because it is cheaper we increased the instance size, which made it faster and improved our iteration times by 2x.
Our main repository saw an average decrease in CI runtime of 8 minutes (23 -> 15). Another repository could fully utilize 16 threads, which dropped its runtime by 15 minutes (20 -> 5).
Simple
RunsOn comes as a CloudFormation stack that you install in your own AWS account. One-click install, one-click upgrades.
It is built using modern tooling (Go), and follows the KISS principles, while blowing away most of the VC-funded competition in terms of price, speed, hardware availability (even GPUs!), and concurrency.
No need to manage a k8s cluster with ARC, or deploy complicated, half-maintained stacks.
Observable
RunsOn automatically publishes metrics for consumed minutes across repository, workflows, runner types, and result (success, failure).
Can be used to monitor the most expensive workflows, detect anomalies, and gain insights into failure rates of your workflows.
10x cheaper, up to 2x faster
Latest generation of AWS instances, fast volumes, and 10+ Gbps network. Your
builds will be faster.
Use the calculator to see how much you can save compared to GitHub runners. Look at the benchmarks to see how much faster your runners can be.
Unlimited concurrency
Tested with hundreds of concurrent jobs at once.
One-line change
1-1 workflow compatibility with official GitHub runners. In fact, RunsOn is the maintainer of the AWS images for GitHub Action, for both x64 and ARM64.
Get the runners your team deserves
Forget statically assigned runner types. Specify RAM, CPU, DISK size, at runtime. Make your runners fit your workflow needs, not the other way around.
Not a SaaS
Cut the middleman. Keep your code and workflow secrets private. RunsOn is installed in your own AWS account.
Read the docs
Learn more in the RunsOn documentation.
4 CPUs, x64, 16 or 32GB ram, default instance family (m7a, m7i):
- runs-on: ubuntu-latest+ runs-on: runs-on,runner=4cpu-linux-x64,ram=16+32,disk=10016 CPUs, ARM64, explicit instance family:
- runs-on: ubuntu-latest+ runs-on: runs-on,runner=16cpu-linux-arm64,family=c7g32CPUs, on-demand only, custom AMI:
- runs-on: ubuntu-latest+ runs-on: runs-on,runner=32cpu-linux-x64,ami=my-custom-ami,spot=falseI've been testing Kubernetes with ARC: more complex, each runner takes ~25 seconds to boot (on an existing node), far less attractive than the simplicity of RunsOn.
CI times are down from ~30 mins to 12 mins, thanks to the larger instances. S3 being faster also helped our one not-very-parallel job.