Skip to main content
Kubernetes with Helm is the recommended way to run Alchemy. Hosts need /dev/kvm, so they run on bare-metal nodes or on VMs with nested virtualization. The server runs anywhere. To run Alchemy on your own machine from source, follow the Quickstart instead.

Helm chart

Every release publishes the chart to oci://ghcr.io/galaxy-io/charts/alchemy. The chart’s version matches the release, and it deploys the server and host images of that same version. Alchemy needs a PostgreSQL DSN, a NATS URL and an encryption key for tenant secrets. Put them in a Secret and point the chart at it:
Without --version, Helm installs the newest release. NATS must have JetStream enabled with file and memory storage and a max_payload above 8 MiB. See Storage. Before the hosts can schedule, label and taint the KVM nodes. The chart’s hosts select and tolerate alchemy.getgalaxy.io/kvm=true by default:
Kubernetes covers the chart in full, including the vendored PostgreSQL and NATS for test clusters, and KVM nodes covers preparing the nodes.

Container images

The images live under ghcr.io/galaxy-io/alchemy: The server and host images are tagged with the release version, such as 0.0.1, and with sha-<commit>. Use the same version for both. The chart defaults their tags to its own app version. The runtime images are published apart from releases, whenever one changes on main. Each carries its language version as its tag, plus that version with a hash of the runtime’s source tree. Point the chart’s runtimes.imagePrefix at a registry of your own to serve the runtimes from there. See Runtimes and builds.

Verify a release

The chart and every published image carry a build provenance attestation. Verify them with the GitHub CLI:

Next steps