Cloud infrastructure, kept small until the workload needs more.

Cloud infrastructure covers a few jobs that should not be confused. A place for the application to run. A database and files that must not live only on a laptop. A pipeline from the repository to a release. An alert that reaches a person when the service stops answering. And the bill. A useful setup names which of those you have and refuses the rest until there is a reason.

The work can be the last mile of a build, a move off a single server, or a takeover of an account nobody can draw. It is a poor use of money when the application itself is the problem and a new host will not change that. In that case the first conversation is about the software.

The default is deliberately small. A web application often runs as a managed front end, a managed database, object storage if there are files, and a pipeline from the repository. Vercel is a normal home for a Next.js application. AWS or Azure is a normal home when you are already there, or when you need a private network, a worker, or a region the other host does not give you. Docker is used when the runtime has to match between the pipeline and production. It is not a requirement for every site.

We do not put Kubernetes, a service mesh, active-active multi-region, a separate observability product, or an idle queue on the first drawing. Each can be right later. None of them is a sign of seriousness on day one. Seriousness is a deploy a second person can run, a bill you can read, and a restore you have tried. A preview the host gives you for free is welcome. A full second copy of production that nobody patches is a cost and a risk.

Setup starts in an account your organisation owns: multi-factor authentication on the owner, a billing alert someone will read, and a short list of who else is allowed in. We then place only what the application needs. The region follows where the users are and any real constraint on where the data may live. If that constraint is still a rumour, we stop and name it. Guessing a region is a migration you pay for twice.

The pipeline builds from the repository and deploys. Logs are kept for a stated period, not forever. A health check hits a path a user depends on, and the alert goes to someone who can act. Backups match how much data you can bear to lose, and we restore once before calling the backup real. Handover is a page: what is running, where, how to deploy, how to roll back, and what the monthly bill should contain. A static site does not get the machinery of a system that holds personal data. If you already have a sensible host, we keep it.

A migration fails when it is secretly a rewrite. The host, the database, and the features all change, and nobody can say which one broke the afternoon. We separate them. First an inventory: what is running, what talks to it, which jobs live in a schedule or in someone's memory, where the secrets are, and what an hour offline means in your words. Then a copy, a rehearsal including a restore and a rollback, and a cutover you can reverse. The redesign, if there is one, comes after the new home is boring.

DNS, certificates, and the jobs that only fire at night are the pieces that get forgotten. So does the old environment, which keeps costing money. We name a date when the old system becomes read-only, and a later date when it is removed. Taking over an undocumented account is a migration even when the servers do not move: draw the map, close the urgent holes, then change the release path in stages. If the application is what needs replacing, we say so. A move will not repair it.

Cloud cost follows what you leave running. Idle environments, a load balancer with nothing behind it, a private network you do not need, logs kept forever, a database sized for a launch that did not happen, and traffic that crosses regions because two services were placed carelessly. None of this requires a cost programme. It requires someone to read the invoice against a list of what should exist.

On day one the account gets a budget alert at a figure you would be unhappy to discover later. Resources are named so an unknown line can be traced. Reserved capacity waits until the shape has been stable. A discount on the wrong size becomes a floor. Spreading the system across three providers to look portable creates three bills. Portability is a copy of the data and a pipeline you understand. Each month the question is the same: which line is this, and is it still doing a job? If nobody can answer, it is a candidate to turn off, after you have checked it is not the backup.

The small stack stops being honest in a few situations. A slow job is blocking requests, and a worker would take it off the request path. The data has a stated residency or contract limit the current host cannot meet. An hour of downtime has a cost you can describe, and you no longer accept a single point of failure. Advisers have required a control — private networking, a firewall, a specific form of encryption, an audit of who opened a record — and the setup cannot express it.

'We might get big' is not one of those reasons. When a reason is real, we add the piece and write down the cost and the owner. The healthcare software guide and the DPDP guide are orientations, not rulings. If advisers have stated a region, that statement is an input. A system that fails because nobody owns the alert needs an owner, not a more expensive host.

Scoping is a short look at what runs, how often it changes, where the data may live, what an hour offline means, and who is on call. From that we write the setup or the migration, and an exclusion list of the platform pieces we are not building. A fixed architecture offered before that look is a template. Templates can explain a default. They are a poor way to price your account.

Sunbeam is a full-service software agency. Infrastructure sits next to the product, the automation, and the website. The same team that builds the application can put it somewhere you can operate, or can take over a deployment someone else started. We work from Lucknow with clients across India and worldwide. The account remains yours if we stop. Talk with Sunny, or read the business automation guide if the workflow is the real project. Technology consulting is the better first step when the decision is not ready for a build.