Graphide CI
How Graphide CI works
Your GitHub Actions jobs on Graphide's runners, tuned by our agents, with the same verdict your own CI gives.
Graphide CI gives you the verdict your CI gives today, sooner. Your workflows stay on GitHub Actions; the jobs run on Graphide’s runners, and every test your job runs on GitHub runs there too.
A failing test always fails the job, exactly as it would on your own CI. Graphide CI changes how long you wait, never whether a job passes.
Three parts
- Our runners. Jobs run on machines we operate, with dependencies and build caches already on the runner when the job starts, and long test steps split across as many runners as it takes.
- Every test runs. Graphide leaves no test out today; test selection is being built and is not on. Which tests run, and why.
- Our agents. They tune how your jobs run (caches, machines, and test splits where a check shows they are safe), all on our end. What the agents change.
Moving a repository
-
Connect GitHub: start your Graphide org’s free trial and install the Graphide GitHub App on your GitHub organization. Graphide CI needs a GitHub organization; its runners do not take jobs from a personal GitHub account.
-
On your dashboard, press Switch to Graphide CI on the repository’s row, or ask your coding agent to switch it. Nothing moves until one of you does.
-
Graphide opens one pull request and merges it. It changes the
runs-online of each job on GitHub’s standard Ubuntu runners or on your own x64 Linux runners to Graphide’s label, an x64 Linux machine with 8 vCPUs and 16 GiB:runs-on: graphideIf a workflow runs actionlint, the pull request also lists the label in
.github/actionlint.yaml. With agent signals on (the default), it adds Graphide’s agent hook files where they are missing and never edits one you have (Agent signals).From the next push, those jobs run on Graphide’s runners. If GitHub holds the pull request back (a required review or check), it stays open: once it is approved, press Merge PR on the dashboard, or ask your agent to merge it. Some jobs keep running where they do today, and the pull request lists them: macOS, Windows and Arm jobs, your own runners for another platform, deploy jobs (a job with
environment:), jobs that publish to npm with provenance, and jobs that ask for more than 8 vCPUs.
To move a repository back, open it on the dashboard (click its name), go to
CI settings, press Switch back to GitHub Actions under Danger
zone, and type the repository’s name to confirm. It gives up Graphide
CI’s speed-ups and test verdicts on that repository. Graphide opens a pull
request that puts each job’s runs-on
back as it was before the switch, and merges it; until it merges, your
jobs keep running on Graphide’s runners. While the switch’s own pull
request is still open, press Close the pull request there instead:
nothing has moved yet. Reverting the switch’s pull request on GitHub
works too.
When a row says “Stays on your own CI”
The row says why, and what would change it. Rows you can act on come first in each organization’s list, and rows that stay come last.
| The row says | Do this |
|---|---|
| Needs a GitHub organization | Move the repository to a free GitHub organization, connect it to your Graphide org, then press Switch. |
| No CI to move (it has no GitHub Actions workflows) | Nothing, unless you want CI there: add a workflow under .github/workflows with a Linux job (runs-on: ubuntu-latest), and Switch shows within a few minutes. |
None of its jobs can move, with a count of jobs on self-hosted or third-party runners, or with a runs-on Graphide cannot rewrite |
If such a job can run on x64 Linux, set its runs-on to ubuntu-latest on the default branch: the row is checked again within five minutes and Switch shows. |
| None of its jobs can move, with a count of jobs on macOS, Windows or Arm | Nothing: Graphide CI runs x64 Linux jobs, so those stay where they run today. |
| A workflow file does not parse | Fix the file on the default branch; the row is checked again within five minutes. |
| Its workflows have no job with a runs-on of its own | Its jobs call a reusable workflow, which decides where they run: switch the repository that holds that workflow. |
| Needs a GitHub setting (the Default runner group blocks public repositories) | GitHub’s Default group keeps public repositories off outside runners. Open the organization’s Settings › Actions › Runner groups, click Default, tick Allow public repositories, then Save. The box shows only once you are inside the group. |
| The App needs more permissions | An owner accepts the Graphide App’s permission request under the organization’s Settings › GitHub Apps, then presses Switch again. |
A row that says Set to run on Graphide CI has jobs that already ask
for runs-on: graphide: there is nothing to switch.
None of these is a lock: whether to change a workflow or a GitHub setting is your choice, and the row says which change would move the repository.
For coding agents
An agent asked to “get me on Graphide CI” does it through Graphide’s MCP
server, https://mcp.graphide.com (Streamable HTTP, OAuth sign-in), with
the person doing only what GitHub makes them do themselves.
- Add the server and sign in. Claude Code:
claude mcp add --scope user --transport http graphide https://mcp.graphide.com, thenclaude mcp login graphide. Codex:codex mcp add graphide --url https://mcp.graphide.com. Other agents: /setup.md. - Call
ci_status. With no installations, the GitHub App is not installed yet: callreview_setup, give the person its install link (GitHub has them click Install themselves), then callci_statusagain. - For each repository,
ci_statusgives its state, and for one that cannot switch amessagein words for the person:can_switch: true: ask the person whether to switch it.state: openedwith nomerged_at: its switch pull request waits on GitHub;ci_switchoffers the merge.state: left: tell the person themessage.
- Call
ci_switchwith the repository. It answers anask: put it to the person, and only on their yes call again with itsconfirm. The answer’snextsays what to do or tell them.
No step needs anyone at Graphide, and nothing in an answer is a lock to get around: if a step needs the person, the answer says which link to give them, and the choice is theirs.
On the runner
- GitHub hands each job the secrets it needs, as it does on its own runners. We use a secret only to run that job, on a machine that is deleted when the job ends.
- GitHub decides whether the job passed, from the job’s own steps, as it does on any runner.
- Treat build logs as you would on any CI: if your build prints a secret, the log holds it.
CI security says where your jobs run, how they are kept apart, and what Graphide keeps from them.
Against GitHub-hosted runners
| Graphide CI | GitHub-hosted | |
|---|---|---|
| Agents tuning how your jobs run | Yes | No |
| Caches warm before the job starts | Yes | No |
| A failing test fails the job | Yes | Yes |
Something wrong or missing? Tell us on Discord or at hello@graphide.com.