Dependencies and runtimes¶
The cloudseed CLI itself needs only Python 3.9+. The tools it drives (Terraform above all, plus cloud CLIs, kubectl and Helm when you use them) can come from three places, and you choose.
| Mode | How | When |
|---|---|---|
| Install locally | cs install <tool> or cs deps install <tool> | the default on a workstation |
| Container | cs deps image, then cs --runtime container <command> | nothing installed locally, or you want isolation |
| Single binary | cs deps bundle builds dist/cloudseed-<os>-<arch> | machines with nothing on them |
When a required tool is missing, setup asks which of the three you want. With -y it stops instead, with exit code 2 and the exact command to run, unless CLOUDSEED_AUTO_INSTALL=1 approved installs up front.
cs deps status # this machine: version, Python, runtime, container engines, and every tool per target
What is required¶
| Tool | Needed for |
|---|---|
| Terraform >= 1.10 | every target (S3 native state locking needs 1.10) |
ssh-keygen, ssh | keys and the bastion |
aws, gcloud, az | optional: login convenience and the cluster commands (fetching credentials). Azure's az login needs az unless you use ARM_* service-principal variables |
| kubectl, Helm | cluster commands and the platform catalog |
| Go 1.25+, qemu-img | VMware: building cloudseed's provider once, converting cloud images |
| VMware Fusion Pro 13+ / Workstation Pro 17+ | the VMware target |
cloudseed authenticates through Terraform's providers: environment variables, AWS profiles and SSO, Google application default credentials and az login.
1. Install locally¶
cs install list # every installable target and whether it is present
cs install terraform
cs install cloud # terraform aws gcloud az
cs install vmware # terraform go qemu-img + the VMware provider
cs install kubernetes # kubectl helm
cs install all # everything above, VPN clients and the agent skills
cs deps install terraform aws # single tools only (its `all` is terraform aws gcloud az)
How tools are installed:
- Homebrew when it is available.
- Otherwise Terraform, kubectl, Helm, Go, k9s and the Databricks CLI come from their official releases, SHA-256 verified, into
~/.cloudseed/bin. Nosudo. awsandgcloudfrom the vendors' installers over HTTPS;azand the Snowflake CLI with pip in their own virtualenv (Python 3.10+).- qemu-img, OpenVPN and Tailscale from the OS package manager.
- kubescape and trivy when you name them, or when
cs scanfirst needs them.
cs install is the one front door and also installs groups, agent skills, agent CLIs and the VMware provider (cs install vmware-provider --rebuild rebuilds it). Installing is always yours: inside an agent session cloudseed refuses it and hands you the command.
2. Run in a container¶
cs deps image # build cloudseed:local (asks Docker or Podman once)
cs deps image --engine podman --rebuild
cs --runtime container setup aws --env dev
cs deps runtime container # make the container the default for every command
cs deps runtime auto # back to local tools, asking when something is missing
The image is built from the repository's Dockerfile with Terraform, the AWS CLI, gcloud, az, kubectl and Helm, on a digest-pinned Python base image. When a command runs in it:
CLOUDSEED_HOMEis mounted at the same path, so every recorded path stays valid;~/.aws,~/.config/gcloudand~/.azureare mounted, andAWS_*,GOOGLE_*andARM_*variables are passed;- tools the container installs live in
~/.cloudseed/container-linux-<arch>/, apart from your host's.
VMware environments always run on your machine (the hypervisor is local), whatever the runtime.
3. Build a single binary¶
dist/cloudseed-<os>-<arch> is one file with the CLI, every Terraform module, the Ansible playbooks, the web console, the skills, the VMware provider sources and a checksum-verified Terraform release embedded. Build it on the same OS and CPU architecture you will run it on; the build uses PyInstaller in a private virtualenv under build/. It is also available as cs install bundle and make bundle.
Cloudseed on the bastion¶
Normal provisioning installs cloudseed and its short alias cs for the SSH login user on AWS, GCP, Azure and VMware bastions. The commands use the exact source payload copied from your current checkout, container or standalone binary; provisioning does not download an unrelated latest Cloudseed release. A standalone binary includes portable Python source for this purpose, so a macOS controller can provision a Linux bastion.
After setup, check the installation from your workstation:
Replace aws and demo with your target and environment. Inside the bastion, cs help and cs doctor work directly. The commands live in ~/.local/bin; login and SSH shell startup include that directory and ~/.cloudseed/bin. An unrelated existing cloudseed or cs command in the installation location stops provisioning with an explanation instead of being overwritten.
When Kubernetes and tools are enabled, the bastion also gets native kubectl, helm and k9s. Downloads use published SHA-256 checksums; re-provisioning retains already installed Helm/k9s clients. kubectl follows the configured cluster minor version when available, including the existing kubeadm default. VMware RKE2 with no configured Kubernetes version uses the upstream stable client and reports that choice; check the running cluster version after authenticating and keep the client within one minor version of the API server.
The bastion has its own local Cloudseed home. Installation does not copy the controller's environment records, Terraform state, private SSH keys, kubeconfig or cloud credentials. Use an authorized bastion identity and kubeconfig for cluster administration; keep infrastructure lifecycle commands on the original controller. Merely installing the CLI does not grant cloud or cluster permissions.
--no-tools keeps the core Cloudseed CLI but skips extra cloud/Kubernetes clients. --no-provision skips this installation, and --sync-only refreshes source files without repairing command links or installing clients. To repair an older bastion from the current controller:
Provisioning checks both command names and the native cluster clients before reporting success. These checks verify the installation; authenticating and reaching the private Kubernetes API remain separate checks.
Related¶
- Installation
cs help deps,cs help install,cs explain dependencies