From catalogue entry to running infrastructure

Five steps. No need to write infrastructure code.

  • 1. Select a deployment

    The catalogue holds 350+ deployment options covering 180+ applications and platform modules. Most applications ship in both a Cloud Run and a GKE variant.

  • 2. Fill form

    Basic mode shows only the variables the module marks as mandatory, and a first deployment is always Basic. The full variable set opens on update, once you buy credits.

  • 3. Review cost and confirm

    The module fee depends on the module's complexity, and is shown on the confirmation screen. Build cost are always metered and charged.

  • 4. Watch build

    RAD runs Terraform as a provisioning job against your project. Build logs stream into the console while it runs, enabling users to view status in realtime.

  • 5. View outputs

    The module's Terraform outputs are published against the deployment. The state is not: it lives in RAD's own Cloud Storage bucket from where updates can be performed.

A deployment in progress in RAD: Prepare Deployment Environment and Initialize and Plan both SUCCESS, Apply Infrastructure Changes WORKING, with raw Terraform apply output streaming beneath it
Step 4, as it happens. The stages report themselves and the Terraform output streams underneath — the same text an engineer running this by hand would read.

The form refuses what the build would have refused

Each form is generated from the module's own Terraform, so what it accepts is what that module accepts. Three checks run before anything is queued.

  • Validation comes from the module

    Where a module declares a rule for a field — a naming pattern, a length limit — the form enforces it as you type and shows the module's own message. You meet it before the build rather than several minutes into one.

  • Quota ceilings are checked

    A value the target project's quota cannot satisfy is refused when the form is submitted, on create and on update. The alternative is an apply that runs for minutes and then stops on a limit that was knowable at the start.

  • Secrets never come back to the browser

    A field the module marks sensitive is masked and written to Google Secret Manager instead of being stored with the rest of the configuration. On update it reads Configured or Not configured; leaving a configured one blank keeps it.

When it fails, you are not left holding a stack trace

The stage that broke is marked, the stages after it stay queued rather than running against a half-built foundation, and the Terraform output that produced the error sits underneath it. The whole log is searchable and downloadable.

  • Search for a fix

    Opens a Google AI Mode search already seeded with that step's own error output, so you are not copying a stack trace into a search box or describing it from memory.

  • Ask for help

    Raises a support ticket with the failure context attached. One route is instant and self-service, the other reaches a person; which you want depends on how legible the error is.

  • Only on a real failure

    A deployment you cancelled yourself does not offer either action, because there is nothing to look up. They appear on a step that genuinely failed.

A failed RAD deployment: Apply Infrastructure Changes marked FAILURE with Search for a fix and Ask for help buttons on the step itself, the Terraform output beneath it, and the remaining steps left QUEUED
The failed step, with its own two actions. The steps above it succeeded, the ones below never ran — and the output that caused it is right there, searchable and downloadable.

And when it succeeds, you can still tell what you got

A row of green SUCCESS marks tells you the build worked. It does not tell you what now exists in your Google Cloud project, and most people deploying through RAD do not read Terraform. One button answers that.

  • Explain this

    Opens a plain-English explanation of the Google Cloud resources the deployment created, and what they cost to run. It works from what the build actually did — Terraform's own added, changed and destroyed counts, and a tally of each kind of resource — not from a guess about what the module usually does.

  • Only the shape leaves your account

    It sends the kinds of resource and how many of each. Your project ID, resource names, service account and user email addresses, and every configured value stay behind. The question is answered by the shape of the estate, so the identifiers never need to travel.

  • One button, not one per stage

    Of the six stages in a build, only the apply has anything to say — the rest fetch code and move archives. A button on every stage would return five shrugs and one answer, so there is one, for the build.

Update, delete, purge

A deployment is not a one-way door. Three actions cover its lifecycle, and one of them is the exit.

Update

Change the module's configuration and re-apply. Terraform plans and applies the difference rather than rebuilding from scratch, so what is already running is left where it is. The build cost of the update is metered.

Changes that destroy

Some fields cannot change without rebuilding what they created, so changing one raises a confirmation before anything runs. An administrator can instead make those fields read-only while a deployment is running.

Delete

Destroys the cloud resources the module created, in dependency order. Resources shared with other module deployments — a project, a network — are retained, so a shared foundation outlives what ran inside it.

Purge

Removes RAD's record of the deployment and its Terraform state, and runs no Terraform. In a project you own you can purge while the infrastructure keeps running; in a RAD-managed project the delete has to finish first, because those resources bill RAD.

RAD's update form for a deployment in a RAD-managed development-tier project: an Enable advanced mode toggle with an estimated 8-credit build cost, above a dialog reporting no configuration changes and offering to re-apply
Reopening a deployment in a RAD-managed project. Advanced mode is offered here like anywhere else, and an update is charged for its build time only, never a second module fee.

What can stop an update before the form opens

An update inherits the state of everything the deployment depends on. Two of those states are handled differently, and the difference is the point.

  • A failed prerequisite refuses up front

    Where this deployment depends on one that failed, was cancelled or timed out, RAD names them in a "Start these first" panel and leaves the submit button disabled until they succeed. Fixing the prerequisite itself is never blocked.

  • A prerequisite still building is parked

    That update is accepted rather than refused, and released once the prerequisite finishes. A dependency in flight is not a fault, so it is not reported as one and you do not have to come back and resubmit.

  • Cancel covers queued and waiting

    A deployment can be cancelled in either state, before it starts building. Cancelling one that is waiting also releases everything waiting behind it: nothing has built yet, so there is nothing to unwind.

RAD applies a retention period: once a deployment has been unused for longer than that period, and its owner has neither signed in within the window nor holds any purchased credits, RAD deletes the record and the Terraform state along with it. A warning email goes out to each user. Deployments in a RAD-managed project are exempt from the sweep entirely, as are the records of the managed projects themselves. Retention removes configuration records: your Google Cloud resources are untouched by it and stay running.

Estate orchestration: deploying several applications as one unit

A solution is a bundle of applications deployed as a single dependency-ordered unit. Take one of the pre-composed ones, which hold between three and eight, or describe what you need and have RAD compose one of up to twelve.

  • Members are released by dependency

    Members without a dependency queued the moment the deployment is confirmed, with three provisioning at a time. A member that consumes another member's outputs waits for the dependencies to be successfully deployed.

  • Outputs become configuration at the boundary

    Where a wiring recipe is defined for a producer-and-consumer pair, the producer's outputs are read and written into the consumer's configuration before it deploys. Other members of a solution take their configuration from the form, and some may need manual configuration following deployment.

  • Teardown runs the graph in reverse

    Destroying a bundle does not fire every teardown at once. A shared foundation is parked behind resources that depend on it. This ordering stops a project being deleted before its members.

  • One reservation for the whole bundle

    Deployment credits for every member are reserved before the first member deployment starts, so a bundle cannot get halfway in and stop for want of balance. Multi-module solutions are discounted.

  • A bundle can be composed from a description

    Rather than picking a pre-composed solution, describe what you are trying to build. RAD reads the catalogue and proposes a combination with a reason for each choice; you accept the ones you want. The saved set deploys through this same pipeline, and RAD wires a pair only where a recipe for it already exists.

The RAD deployments list: a Project GCP module WORKING, three application modules WAITING on it, and older deployments showing SUCCESS with their build durations
One solution, mid-flight. The foundation is WORKING, the members that need it are WAITING, and everything already built carries the time it actually took.

Producer-and-consumer pair wiring

Wiring is defined per producer-and-consumer pair. Additional configuration may be required after the deployment finishes. Examples are listed here:

Where it deploys

Two targets, and the catalogue and engine are identical for both.

Google Cloud project you own

RAD deploys as a principal inside your Google Cloud project, subject to your organisation's policies and unable to widen them. Every Google Cloud region is available to you; region narrowing is a property of RAD-managed projects only. RAD holds and manages the deployment configuration, and the Terraform state used to deploy into your project is stored in RAD's own Cloud Storage bucket rather than in yours.

Project RAD creates and governs

If you have no Google Cloud account, RAD creates a project for you in one of four tiers, each in its own Google Cloud folder with its own organisation-policy set. Creating one requires a minimum purchased credit balance for that tier, and you may hold one managed project per tier at a time. The project's Google Cloud spend is recharged to your credit balance, as Google reports it.

Four tiers in four folders

Separate folders are the point: a relaxation made for one tier cannot drift into another. Lab carries the sandbox policy set deliberately, in a folder of its own, so a cohort cannot reach into a customer's own sandbox.

Project tiers

  • SandboxSelf-selectable
  • DevelopmentSelf-selectable
  • ProductionSelf-selectable
  • LabNot selectable; follows a cohort

Regions in a RAD-managed project

  • United Statesus-central1
  • Europeeurope-north2
  • Mexiconorthamerica-south1
  • Middle Eastme-west1
  • Africaafrica-south1
  • Asiaasia-east1
  • Australiaaustralia-southeast2
  • South Americasouthamerica-west1

One region in each of eight geographies. Each was picked as the cheapest available region in that geography, measured from the Cloud Billing Catalog API rather than chosen by reputation. africa-south1 is the only Google Cloud region on the African continent. In your own project none of this applies — you keep every region Google offers.

What the guardrails actually do

Mechanisms, not adjectives. These are the controls applied to the sandbox and lab tiers, which carry the tightest set.

Development relaxes three things

The same credential-hygiene floors apply. An external IP is permitted only where OS Login and Shielded VM are in force; a publicly readable bucket only where uniform bucket-level access is on; a public database IP only through the Cloud SQL Auth Proxy, because authorised networks are blocked outright. Developers can configure enabled Google Cloud services.

Production inherits those ceilings

Production is not the uncapped tier. It takes development's quota set wholesale: 48 Compute Engine vCPU and 32 Cloud Run vCPU per region, 32 GB of Memorystore Redis per region, 1 TiB of BigQuery scan per day. A value above one of those ceilings is refused when you submit the form, naming the ceiling it exceeded. Quotas can be increased on request.

Lab is for cohorts, early access

Nobody picks the lab tier from a list — it appears on no deploy form. It follows from deploying for participants: attach a roster, choose who the deployment is for, and each participant's project lands in lab. Only trainers and administrators can do that, and each deployment's credits are debited from that participant's own balance, never the trainer's.

Governance the platform enforces

These access rules are enforced by the server on every request.

  • Roles are resolved on the server

    What you are allowed to do is read from your stored user record on the server, on every request — never from anything the browser sends. A modified client asks the same questions and gets the same answers.

  • Support access is tied to an open ticket

    A support agent can see your deployments only while a ticket of yours is assigned to them and still open. Resolving or closing it ends the access — at once in the console, and within five minutes on the API; reopening restores it.

  • Secrets are outside support's reach entirely

    Raw configuration values and deployment outputs are excluded from support access, as is anything destructive — support cannot tear down what it can see.

  • Money-bearing actions leave a record

    Credit grants, adjustments and revenue splits write an audit record, and cross-tenant support reads of a deployment's status, logs or credit history are recorded too.

A governance view showing roles, scoped access and audit records

Spend safety, described precisely

Three separate mechanisms that are often collapsed into one word. They do different things and one of them does less than people assume.

Build cost is metered

You are charged for the resources needed to perform the deployment, measured from its real duration at a published per-hour credit rate. The cost is deducted once the build finishes, so you pay for the time it actually took rather than an estimate. The module fee is separate, and is shown for review before you confirm.

Monthly spend alert

Each tier carries a monthly Google Cloud budget that raises alerts as spend approaches and then exceeds it. Those alerts reach our administrators and finance team, not the project's user, because the spend lands on our billing account. A budget notifies; it does not refuse an API call, and raising one permits no extra spend.

Billing suspension

RAD is designed to disable billing on a managed project when the account's purchased credit balance goes negative, and to restore it once the balance is topped up. It checks on a fifteen-minute schedule, and only ever reverses a suspension it applied itself — never one you or Google put in place.

Scope, and where RAD is the wrong answer

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

One platform, four kinds of team

Everything above is the same whoever you are. What it is worth depends on what you deploy and how often — pick the closest fit.

Platform teams

One reviewed catalogue deployed into projects your organisation already governs, as ordinary Terraform.

Platform teams

Startups and hubs

The estate a product needs, in your own project or in a governed one RAD creates if you have no GCP account.

Startups and hubs

Training organisations

A real project and a real deployment per participant, provisioned for a cohort in one action. Early access.

Training

Module developers

Make your software deployable on Google Cloud from one form, or publish and deploy your own Terraform modules.

Module owners

Deploy one module into one of your own projects

It is the fastest way to check any of this for yourself. Signing up gives you 300 credits and does not ask for a payment method.