GitHub Actions up to 4× faster.
In your AWS, for up to 80% less.

Runners are in early access. Request access and we'll set you up.

When a job asks for a Ravion label, Ravion boots a Firecracker sandbox in your AWS account and hands it the job. It's gone when the job finishes.

No. Jobs check out and build your code on machines in your own AWS account, and caches are stored in S3 in that account. Job logs go straight to GitHub, as with any runner. Ravion sees job details like the repo, branch, and status, and uses them to pick a machine for each job.

Jobs marked interruptible=true can use Spot. If AWS reclaims the host, Ravion moves the running job live to another host, so it keeps going without a restart. If a move can't finish before AWS's deadline, the job restarts on another machine, and its last attempt runs on on-demand.

Yes. actions/cache works unchanged. Whole-disk caching follows your GitHub cache settings by default, bringing back package caches, Docker layers, and build caches. Use disk_cache=off for a fresh disk, or set disk_cache=auto or a cache key to override. actions/checkout cleans untracked files by default, so keep caches outside the checkout folder.

Turborepo, Bazel, Go, Gradle, Rust with sccache, and C and C++ with ccache share one remote build cache in S3 in your own AWS account, with no workflow changes. Ravion points each tool at it for every job, so a job replays outputs that another job, another branch, or a fresh disk already built. Gradle needs its build cache turned on. Remote caches you already configure, such as Vercel Remote Cache or your own Bazel remote, keep working.

Yes. Agents open more PRs, and every PR runs CI. Whole-disk caching keeps each run short, and Spot keeps the volume affordable.

As much as we can. If GitHub's webhooks run late or go missing, Ravion checks GitHub for queued jobs every minute and starts runners for them, and a sweep every five minutes catches anything else. Caches are served from S3 in your own AWS account, not GitHub's cache service, so restores keep working when GitHub's cache is slow or down. GitHub still assigns every job, so if GitHub Actions stops handing out jobs entirely, no runner can start one, and Ravion picks them up as soon as it recovers.

It works until the machine sleeps, fills its disk, or one bad dependency gets access to everything on it. Ravion gives you the speed and price of your own hardware, with each job in a fresh, isolated sandbox in your AWS account, autoscaled, retired, and replaced automatically.

Savings come from lower compute rates and fewer billed minutes. On real CI workloads run side by side, that adds up to as much as 80% less than GitHub-hosted runners, 65% less than Depot, and 50% less than Blacksmith. You pay AWS directly for on-demand or Spot compute, with no markup from us. Whole-disk caching skips repeated installs, image pulls, and builds, so jobs finish sooner. Our fee is $0.002 per minute, and Team includes 50,000 minutes a month. Actual savings depend on the workload, instance type, and capacity mix.

On real CI workloads run side by side, jobs finish up to 4× faster than GitHub-hosted runners, up to 2× faster than RunsOn, and up to 30% faster than Blacksmith and Depot. Each job gets a Firecracker microVM on current-generation EC2. Whole-disk caching brings back package caches and Docker layers from the last successful run. Actual speedups depend on your workload.