Graphide CI
10× faster CI, kept fast by AI agents
Your GitHub Actions jobs run on our runners. A team of agents keeps every job fast as your code changes, on every repo.
Backed by Y Combinator47× faster1
Change one line, and agents keep your CI fast
Change one line
Point runs-on at Graphide. Nothing else changes.
Run on our runners
Warm machines, sized to each job.
Agents keep it fast
Every job retuned as your code changes.
Ten long open-source CI jobs, timed on both1
| Job | GitHub | Graphide CI | Faster |
|---|---|---|---|
| aquasecurity/trivyBuild Test | 23m 33s | 30s | 47× |
| coredns/corednsTest Makefile.release | 17m 42s | 24s | 45× |
| tokio-rs/tokiomiri-test | 19m 58s | 50s | 24× |
| rclone/rclonebuild (other_os) | 4m 24s | 11s | 24× |
| bytecodealliance/wasmtimeMiri (wasmtime-cli) | 21m 42s | 1m 14s | 17.6× |
| bytecodealliance/wasmtimeMiri (component) | 11m 32s | 46s | 15.1× |
| rust-lang/futures-rscargo miri test | 10m 35s | 52s | 12.2× |
| rust-lang/hashbrownmiri | 6m 56s | 36s | 11.7× |
| microsoft/playwrightMCP (firefox) | 29m 54s | 2m 39s | 11.3× |
| webpack/webpacktest262 | 27m 45s | 2m 35s | 10.7× |
Every job runs in its own sandbox
How we keep jobs apartOne machine per job
Fresh for every job, destroyed after.
Walled off
No reach to other jobs or private networks.
Secrets stay with the job
They reach only the machine running the job.
Built by Dan McGee and Benjamin McIntyre in San Francisco
Questions
Does Graphide CI ever change whether a job passes?
No. If a test fails on our runners, the job fails.
How do we start?
Book a call, or install the Graphide GitHub App and press Switch to Graphide CI on a repository. That opens one pull request that changes its Linux jobs’ runs-on line.
What does it cost?
Graphide CI is one fixed monthly price, quoted from what you run today. PR review is $1 per pull request after the first 10. See pricing.
Where does our code go?
We read only the repositories you turn on. We never train a model on your code and never sell it. Privacy policy.
One fixed monthly price for your CI
Try it free for 7 days or $10 of CI usage, whichever comes first. Nothing is charged.
How we measured
Ten long open-source CI jobs, timed on GitHub-hosted runners and on Graphide CI
Our runs were made on November 5, 2026, by hand, on our own copy of each project, with changes to how the job runs and none to what it tests. We picked each job’s warm machine image ourselves, and ran on 64 or 128 vCPUs; GitHub’s standard hosted runner for open-source projects has 4, and a customer’s runners on Graphide CI go up to 32 today. GitHub’s time is its own run of the same job: the median of its recent runs or its run at the same commit, and for rclone its fastest run with a warm cache. Steps left out on our side: trivy’s disk clean-up (also taken off GitHub’s time), rclone’s upload, library install and two cache steps, and webpack’s Codecov upload (GitHub’s 10-second median for it is added to ours). Ours is runner start to finish, without time in the queue. Every test name and result (for a build job, every target built) was compared with GitHub’s log of the same job. Eight of the ten are Rust Miri or Go jobs, and how much faster your own job gets depends on what it does.
Known bugs caught by Graphide review and by Greptile, on the same pull requests
74 real pull requests, 68 of them from one open-source project (typescript-go), held out while we tuned the reviewer, carrying 36 known bugs. Graphide review as it ships today ran on them once on October 26, 2026; Greptile’s reviews of the same pull requests were judged on October 24, 2026. The same grader, which was not told which tool wrote what, matched each tool’s findings to the known bugs and counted its false alarms on the pull requests with no known bug.

