The same landing zone, authored six times

Nothing here is anyone's fault. It is what happens when six squads each need Google Cloud infrastructure in the same quarter and there is no shared way to get it.

Duplication

The same service accounts, the same network, the same secret plumbing, written again by a team that could not find the last version or did not trust it.

Divergence

Six copies that started identical and are now six different answers to the same question, each one a separate thing to patch when something changes.

A review that never converges

Security reviews the same pattern repeatedly because each squad brings its own variant. The review effort scales with the number of teams, not with the number of patterns.

No single answer

What is deployed, who deployed it, and what did it cost — three questions with three different sources of truth, none of them complete.

One catalogue, your projects

RAD is a catalogue of ready-made Google Cloud deployment options and an engine that provisions them. The engine does not host anything — it writes Terraform into a project you nominate.

  • 354 deployment options across 190 applications and platform modules

    Counted 2026-08-17. Most applications ship in both a Cloud Run and a GKE variant, so the operating-model choice stays yours rather than being made for you by the catalogue.

  • Subject to your org policies, like any other principal

    When RAD deploys into your project it runs inside your organisation's policy boundary. It cannot widen that boundary, and a deployment that your policies forbid fails the same way your own pipeline would fail.

  • Every Google Cloud region stays available

    Region narrowing applies only to projects RAD creates and manages. In your own project you keep the full set of regions Google offers, and your existing residency rules apply unchanged.

  • A form, not a fork

    Basic mode asks only for the variables that are genuinely mandatory. Advanced mode exposes the full variable set of the underlying module for the cases where a default is wrong for you.

A platform engineering team reviewing Google Cloud infrastructure

It is your Terraform, in your project — and RAD holds the state

This is the section to read first if the question in the room is "why would we let a third party near our cloud?" — and it is the section where we tell you the one thing you do not get, before your reviewer finds it.

What RAD writes

  • ArtefactStandard Terraform
  • State locationRAD's Cloud Storage
  • Project ownerYou
  • Cloud scopeGoogle Cloud only

What RAD is not

  • In the request pathNo
  • A hosting providerNo
  • A proprietary abstractionNo
  • Required to keep runningNo

If you stop using RAD

  • Your workloadsKeep running
  • Your resourcesStay in your project
  • The Terraform stateStays with RAD
  • Ongoing changesYour own tooling

Read that third row before your security reviewer does. Every deployment's Terraform backend points at RAD's own Cloud Storage bucket, in RAD's project — not at yours — and there is no state download in the product today. So the accurate statement of ownership is narrower than the usual one: the Google Cloud project is yours, the resources are yours, the module source is ordinary Terraform, and the state file that tracks those resources is held by us. If you leave, what you keep is the running infrastructure, in your own project, under your own policies; what you would have to rebuild is the state. For a team whose standard is that every production resource is under its own state file, this is a real objection and the answer is "not today", not a reframing.

RAD does not hold it indefinitely either, and that cuts the other way. After the retention period, a deployment record that has gone untouched in the platform, whose owner has neither signed in during that window nor holds purchased credits, is deleted along with its state — after a warning email that must actually have been sent, or the record is kept. Your resources are not touched by that; only RAD's record of them is.

Traffic never passes through RAD

RAD provisions infrastructure and then steps out of the way. Nothing your users touch resolves to us, so an outage on our side is not an outage on yours.

What leaving actually looks like

The resources stay where they are and keep running: they were created in your project, under your policies, by ordinary Terraform. Nothing has to be migrated and nothing stops working. What does not come with you is the state file — it stays in RAD's bucket — so a team that wants to keep managing those resources as code re-adopts them into a state file of its own. That is a real piece of work and we would rather quote it honestly than call the exit free.

Diligence reads ordinary infrastructure

Whoever inspects your estate next — an internal audit, an acquirer's technical review, a new platform lead — reads Terraform they already understand rather than learning our abstraction first.

It deploys estates, not single applications

The catalogue also contains 62 pre-composed solutions, defining 291 member deployments drawn from 151 distinct applications across 12 categories. Counted 2026-08-17.

Dependency-ordered, not stage-gated

A solution averages 4.7 applications, from three to eight. Members with no prerequisite are queued on confirmation and provision three at a time; a member that consumes another member's Terraform outputs waits for that specific deployment, not for a stage in front of it.

Eight wired pairs, and no pretence about the rest

For eight producer-and-consumer pairs, RAD is designed to copy the producer's real Terraform outputs into the consumer's configuration before it builds — Elasticsearch into RAGFlow and Zammad, Ollama and LiteLLM into OpenWebUI, ClickHouse into Plausible, Synapse into Element, Gitea into Woodpecker, Directus into Meilisearch. Those eight pairs appear in eight of the 62 solutions; in the other 54, every connection between members of a solution is integration work your team still does, and the automatic part is a beta capability.

Teardown runs the graph in reverse

A shared project or a shared service is destroyed after the things living inside it have gone, not before. Getting this ordering wrong is what turns a clean teardown into a set of failed destroys and orphaned resources.

Or a governed project RAD creates, in one of four policy tiers

Bring-your-own project is the default for an organisation that already has Google Cloud. The alternative exists for the teams inside your organisation that do not — a proof of concept, a training cohort, an external contributor. Each tier is a separate Google Cloud folder with its own policy set, so one tier's rules cannot drift into another's.

Sandbox

  • External IP addressesDenied
  • Default networkSuppressed
  • Serial port accessDisabled
  • Service-account keysCreation and upload blocked
  • Public bucketsPrevented
  • Public database IPsRefused
  • Enabled servicesAllowlist
  • Who can select itAny user

Development

  • Credential hygieneSame floors as sandbox
  • External IPRequires OS Login and Shielded VM
  • Public bucketRequires uniform bucket-level access
  • Public database IPRequires authorised networks
  • Who can select itAny user

Production

  • Policy setGoverned
  • Resource ceilingsDevelopment's set, raisable per project
  • Compute Engine vCPU per region48
  • Cloud Run vCPU per region32
  • Memorystore Redis per region32 GB
  • BigQuery scan per day1 TiB
  • GPUs and Vertex AI acceleratorsZero
  • Monthly spend alertNotifies only; never blocks
  • Who can select itAny user

Lab

  • Policy setTightest guardrails
  • Intended useTraining cohorts
  • Who can select itTrainers and admins only
  • Feature maturityEarly access

The production tier is governed and quota-bounded, not uncapped: it takes development's ceilings wholesale, because the tier that can spend the most had the least standing between it and the spend. The consequence is one your architects should weigh before choosing it — a workload that legitimately needs more than 48 vCPU or 32 GB of Redis in one region is refused rather than served, and a refusal in production is an outage. Every one of these ceilings is raisable per project on request, which is the handle an uncapped tier would not give you. None of this applies to a project you own: your own quotas are yours.

Eight regions, one in each geography

A RAD-managed project deploys into one region in each of eight geographies: us-central1, europe-north2, northamerica-south1, me-west1, africa-south1, asia-east1, australia-southeast2 and southamerica-west1. africa-south1 is Google Cloud's only African region. Each was selected as the cheapest available region in its geography, measured from the Cloud Billing Catalog API.

347 of the 354 module options are eligible for a managed project

Seven opt out — they are named on the catalogue page with the reason. Everything else can be deployed either way.

Creating a managed project requires a verified email and a minimum purchased credit balance

The minimum is set per tier, appropriate to what that tier can do. Free monthly credits do not satisfy it.

Each managed project carries a monthly Google Cloud spend alert

An alert notifies as spend approaches and exceeds the figure. It does not block an API call, and we do not describe it as though it does. The enforcement mechanism is separate and described below.

Separation of duties, enforced by the platform

We hold no security certification or third-party audit attestation, and we will not imply one. What we can do is describe the mechanisms precisely enough for your reviewer to judge them.

Chargeback that reconciles

Every deployment is charged against a credit balance, and every charge is a row you can read and export. Credits are the unit: 10 credits = $1.

A module fee, shown before you confirm

Between 40 and 300 credits depending on the module's complexity, and nothing at all for nine of the 354 options, which include the migration, service-mesh and reference-architecture modules. The figure appears in the confirmation step, before anything is provisioned, alongside the balance it will come from.

A build cost, metered from real time

Charged from how long provisioning actually ran, at a published per-hour credit rate. More than half of recorded deployments finish inside 20 minutes, and more than nine in ten inside 30. A failed build carries no module fee.

An exportable ledger

Credit history is per-transaction and exportable, so a platform team can attribute spend back to the team or project that caused it rather than reconstructing it from cloud billing alone.

One reservation across a solution

A multi-application solution reserves its cost as a single sum across every member, with a bundle discount of 15% on three to four modules, 20% on five to six and 25% on seven or more.

Publish your own internal modules into the same catalogue

The catalogue is not a closed set. The same publishing path that brings a partner's module in works for a module of your own.

Your repository, your Terraform

You connect a GitHub repository containing your own modules. RAD reads the module's variable definitions and renders the same guided form your teams already use for catalogue modules.

One interface, one audit trail

An internal pattern and a catalogue module are deployed the same way, charged the same way and recorded the same way — which is the point, because two systems produce two answers to "what is deployed".

Internal-only visibility: ask us

If your requirement is that a published module is visible to your organisation and to nobody else, raise it with us as part of an evaluation rather than assuming it. We would rather confirm the exact behaviour with you than promise it on a web page.

Where RAD is the wrong fit

Better to read this now than to discover it in week three of an evaluation.

Google Cloud only

Every module targets Google Cloud. If your requirement is one console spanning several cloud providers, RAD is not that, and governance spread that thinly would not be worth having.

Terraform is the contract

If your organisation has standardised on a different infrastructure language and will not accept Terraform state as an artefact, the catalogue has nothing to give you.

It is in beta

The catalogue, the engine and the documentation are built and running. Newer features — cohort provisioning and trainer roles, shipped in August 2026 — are early access, and we label them that way.

Not the cheapest way to run one application

RAD earns its keep across many deployments and many teams. For a single small application that one engineer will maintain forever, writing the Terraform yourself is cheaper and we will say so.

How to evaluate this in a week

The evaluation that settles the argument is not a demo. It is one module, into one project you already own, reviewed by the person who would have to sign it off.

What to read next

Everything a reviewer usually asks for is on this site or in the documentation. There is no gated pack, and nothing here is behind a form.

How it works

The deployment path end to end, estate orchestration, and where RAD deploys.

Platform

The catalogue

354 deployment options, 190 applications and platform modules, and the seven that cannot use a managed project.

Modules

Documentation

346 hands-on labs, 523 module configuration guides and 41 certification study guides across 7 Google Cloud tracks.

Open the docs

The objections

Ownership, lock-in, what a spend alert does not do, who can see your deployments, and whether RAD is production-ready.

FAQ

Start a technical evaluation

Tell us which project you want to deploy into and which pattern your teams keep rewriting. We will help you get one module into one of your own projects this week.

RAD is in beta. The platform, catalogue and documentation described here are built and running; features marked as early access are newer and we say so where that is true. Catalogue figures counted 2026-08-17.