A control plane for CI and infrastructure:
the process and the resource in one model.
Graphene is for automation that must manage both work and the infrastructure where it runs. A pipeline is a typed Go program. Machines, containers, Kubernetes objects, files, and artifacts are durable resources with state, an owner, and a lifetime.
That shared model solves the gap between CI, which normally sees only steps, and IaC, which normally sees only desired state.
What changes in practice
| Need | Graphene model |
|---|---|
| Create an environment for one test run | The run owns the resource tree; terminal completion finalizes it |
| Keep a failed environment for investigation | Transfer its subtree to the pipeline stand, optionally with a TTL |
| Work on a remote or existing machine | Its agent connects outward; actions need no inbound control port |
| Reuse data between pipelines | Publish and attach a durable artifact record |
| Recover from a worker crash | Execution resumes against recorded resources instead of recreating them blindly |
| Explain a failure | Read the same record's state, events, logs, metrics, and traces |
Resource libraries bring typed implementations without changing this model. The current libraries cover Docker, Kubernetes and Crossplane, Git, and files.
Choose a path
| If you want to… | Start here |
|---|---|
| Understand the product and its current boundary | Current state → What you can build |
| Try it without running a server | First local plan |
| Run a complete local workflow | First development run |
| Write a pipeline | Pipeline binary commands → Main → Core SDK reference |
| Operate from the command line | graphenectl |
| Operate visually | Graphene Studio |
| Integrate another client | Management API |
| Configure or diagnose an installation | Operations |
Current boundary
The repository provides a complete development installation, not a hardened production distribution. Go is the only built-in pipeline authoring API. Pkl/KCL and a language-neutral YAML pipeline format are not current product surfaces. Exact shipped and unfinished capabilities are kept in Current state.