Hey everyone, we're Dan and Ben, cofounders of Graphide. We've been building things together since middle school in Cincinnati, and today we are launching our CI runner!
Engineers now run fleets of coding agents, all opening and merging PRs at once. Every one of those PRs waits on CI, often several times as the agent pushes, fails, fixes and pushes again. Several customers have told us that before swapping to our runner, a basic feature with a 30 minute CI run would sit in that loop for upwards of 3–6 hours. With fast CI, that same team can now build an entire product in a few days.
TL;DR: Graphide is a GitHub Actions runner that combines hardware and software fixes to make your CI faster than either can alone. On our benchmarks, 16 of 24 CI jobs finished in under a minute, oftentimes like in Trivy's case going from a 25 minute build to sub 30 seconds, with identical results.
Is CI slowing down your agents? Switch to Graphide at graphide.com. Onboarding takes 2 minutes.
The benchmark
We ran 24 real CI jobs from open-source projects on Graphide, including 7 from our F26 batchmates, and compared each against that project's own recent GitHub runs. Every job ran its full test suite, with no tests skipped. 16 finished in under a minute, and 19 ran 10x+ faster, up to 50x.
Our F26 batchmates
Our batchmates repos are newer, so their CI starts out shorter, but it still got 6–18x faster!
GitHub times are the median of each project's recent runs. Graphide times are the median of 5 runs.
How we get there
Your CI time is its critical path: boot + install + compile + your longest test lane. Faster hardware shortens each step a little. We go after the whole path:
We work by starting with plain runners, and then applying our speedups over the course of the day in order to make sure nothing breaks with your CI. Every speedup also goes into a shared library, so each new repo starts faster than the last.
Optional: pruning. All the numbers above run every test. If you turn pruning on, we map which tests each file can affect and skip the ones your change can't reach. That cuts costs by roughly another 50-70%.
Security
Every job runs in its own microVM, created for that job and destroyed when it ends. Nothing carries over between jobs, repos or customers.
Pricing
Pay as you go at $0.003 per vCPU-minute, we also offer fixed plans with unlimited usage if you reach out to us! (also comes along with our PR review bot) See pricing.
Team
Our ask
Questions about how it works? Email us or reach out on LinkedIn!