Skip to content

FAQ

The basics

What is cloudseed, in one sentence?

A command-line tool that builds secure landing zones (network, hardened bastion, logging, security baseline) on AWS, Google Cloud, Azure and local VMware with Terraform and Ansible, then adds a Kubernetes platform, backups, chaos engineering, compliance scans and cost reports on top, with a web console, an MCP server and optional AI agents that all run the same commands.

Is it free?

Yes. cloudseed is open source under the Apache-2.0 license. You pay your cloud provider for what it creates, and nothing else. The VMware target costs nothing beyond your own machine; VMware Fusion Pro and Workstation Pro are free.

Does it send telemetry or phone home?

No telemetry. The network calls it makes are the ones you would expect: your cloud's APIs (through Terraform and the cloud CLIs), downloads of tools, cloud images and Helm charts when you ask for them, and a public-IP lookup (checkip.amazonaws.com and fallbacks) so the bastion can be locked to your address. The web console loads nothing from a CDN.

How is this different from writing Terraform myself?

You could build all of it by hand, and cloudseed's Terraform is right there in terraform/ to read. What you get on top: secure defaults decided for you, a plan and an approval before every change, per-environment working directories with non-overlapping networks, remote state bootstrapped and hardened, Ansible hardening after every apply, "already exists" handling that only adopts what is provably yours, targeted destroys, an undo journal, an audit trail, and a platform catalog with pinned versions. And the same shape on your laptop with VMware.

Can I use it without VMware or a cloud account?

Yes, to look around: cloudseed setup aws --env demo --dry-run renders and validates the complete Terraform with no credentials, and cloudseed finops estimate aws --env demo prices it. To build something real you need either VMware Fusion / Workstation or a cloud account.

Cost

What does a landing zone cost to run?

cloudseed's own estimate for the defaults, at on-demand list prices, before data transfer:

Target About Mostly
AWS (us-east-1) $58 / month NAT gateway ($32.85), bastion, public IPv4, GuardDuty
Azure (eastus) $55 / month NAT gateway ($32.85), bastion, public IPv4, Log Analytics
Google Cloud (us-central1) $11 / month the e2-micro bastion and a static IP
VMware $0 your machine

Managed Kubernetes, a VPN host, bigger nodes and platform items with cloud storage add to this. Run cs finops estimate <cloud> --env <name> for your configuration, and cs finops cloud for the actual bill. See FinOps.

How do I make sure I stop paying?

cs destroy <cloud> --env <name> removes everything the environment created; add --purge-state to remove the state bucket too. cs list shows every environment you have. A few account-wide AWS settings (the S3 public-access block, EBS encryption by default, the IAM password policy) stay in place after a destroy because they protect the whole account; they cost nothing.

Security

Where do my cloud credentials go?

Nowhere new. cloudseed uses your existing logins through Terraform's providers, or the optional local vault (~/.cloudseed/credentials.json, mode 0600). Nothing in ~/.cloudseed is copied to your hosts, no secret is stored in Terraform state or outputs, and logs are redacted. See Credentials.

Is it safe to let an AI agent run it?

It is designed for it, with limits you should know. Agents never receive your credentials (they are stripped from the agent's environment and handed only to child cloudseed commands), every line of output is redacted, human-only commands are refused, and changes wait for your approval (the built-in agent) or for confirm=true (MCP). Claude Code runs with cloudseed's secret files denied. Codex, Gemini and Grok have no such file rules and could read files your user can read, so prefer the built-in agent or Claude Code for sensitive accounts. See Agentic mode.

Who can reach the bastion?

Only the IPv4 addresses you allow (your detected public IP by default), with key-only SSH. 0.0.0.0/0, ranges wider than a /8 and lists covering more than two /8s are refused. Workloads have no public address at all.

Are the web console and MCP server exposed to the network?

No. Both listen on 127.0.0.1 only, require a token, and check the Host and Origin headers. The MCP server also runs over stdio with no port at all. Do not put them behind a tunnel or public proxy.

What it does not do

cloudseed is deliberately focused. Things it does not do today:

  • Multi-account organizations. It manages one account, project or subscription per environment. It does not set up AWS Organizations, Control Tower, GCP folders or organization policies, or Azure management groups.
  • Deploy into an existing network. It creates its own VPC or VNet. It adopts only resources whose tags prove they belong to the environment, never an arbitrary existing VPC.
  • Some logs and policies. Azure NSG / VNet flow logs and GCP organization policies are not managed yet; GCP's default network is left untouched.
  • AWS China and isolated regions. Not supported.
  • IPv6 allow-lists. --allow-ip takes IPv4 only.
  • Spot or preemptible nodes. Clusters use on-demand capacity; spot is a change you make outside cloudseed.
  • Upgrade local Kubernetes nodes. On VMware, kubernetes_version pins what new nodes install and never upgrades a node that is already installed: upgrading a running RKE2 or kubeadm cluster is up to you.
  • Team collaboration. Remote state gives locking, but an environment's configuration and SSH keys live in the working directory on the machine that created it. Sharing environments across people is not a built-in workflow.
  • Windows as a first-class host. VMware detection is experimental. Native Windows environment changes are refused because supported environment locking and a native Ansible control node are unavailable.
  • Public load balancers for apps. Gateways get private load balancers; reach them over the VPN or the bastion.

Supported versions

Component Supported
Python 3.9 or newer (standard library only)
Terraform 1.10 or newer (S3 native state locking)
Terraform providers hashicorp/aws ~> 6, hashicorp/google 6.x and 7.x, hashicorp/azurerm >= 4.65 and < 5
Host OS macOS and Linux; native Windows environment changes unsupported
VMware Fusion Pro 13+ (macOS, Intel and Apple silicon), Workstation Pro 17+ (Linux); older releases are refused for new environments
VMware guests Ubuntu 24.04 (default), Ubuntu 22.04, Debian 12 (amd64 and arm64)
Cloud bastions Amazon Linux 2023 (AWS), Debian 12 (GCP), Ubuntu 24.04 (Azure); Ubuntu Pro FIPS images in FIPS mode
Kubernetes EKS, GKE, AKS; RKE2 (stable channel by default) or kubeadm (1.35 by default) on VMware
Helm Helm 3 and Helm 4
Go 1.25+ (only to build the VMware provider)

Everything else

Can I change the Terraform?

Every stack variable is settable with --var name=value (cs help variables <cloud> lists them). For deeper changes, the stacks under terraform/<cloud>/ are plain Terraform: fork them. The per-environment root cloudseed renders (main.tf.json) is regenerated on every run, so edit the inputs, not the rendered file.

What if I delete ~/.cloudseed?

Don't, while environments exist. It holds each environment's configuration, SSH keys and (for local state and VMware) the Terraform state. Destroy your environments first (cs list shows them), then remove it.

How long does a setup take?

A VMware lab takes about ten minutes the first time (image download and provider build) and a couple of minutes after that. A cloud landing zone takes a few minutes; a managed Kubernetes cluster adds the time your provider needs to create it, often 10 to 20 minutes.

How are the docs and scenarios tested?

The reference is generated from the code by scripts/gen-docs.py. The VMware scenarios run live on VMware Fusion and the agent and console ones on a local machine; the AWS, Google Cloud and Azure scenarios are verified with --dry-run (Terraform render and validate) because the test machine has no cloud credentials.

Where do I report a bug or a security issue?

Bugs and ideas: GitHub issues. Security issues: follow the security policy instead of opening a public issue.