Skip to content
Sweet Gum.
How we build
Our stack · Engineering

Our Web Stack: Next.js, Astro and Static Sites

The tools we build sites with, why we pick the simplest one that fits, and the performance lessons behind it.

Our web stack is deliberately plain: Next.js when a site needs routing and logic, Astro when it is mostly static pages, and hand-written HTML when that is the honest answer. Sites deploy from git to Vercel or Netlify. Fast pages and fewer moving parts are the goal.

Most owners never see any of this, and should not have to. But it affects how fast your site loads, how much it costs to host, and how safely changes go live. Here is what we use and why.

Who this is for

Owners who want to know what their site is built on, or who were burned by a bloated site that loaded slowly. If you are shopping for a new site, our small business websites page shows what we deliver. This page shows how.

Picking the right tool

Next.js

A React framework for sites that need routing, content files, forms with logic, or a lot of pages. It is our default for web apps and for sites with a content system, including this one.

Astro

Astro builds static pages with almost no JavaScript sent to the browser. We prefer it for the "many similar sites" case, like local service sites built from one template. One template, many sites, each one fast.

Plain static HTML

Sometimes the right call is no framework at all. A five to ten page attorney or insurance agent site can be hand-written HTML and CSS with a small amount of JavaScript. No build step, nothing to upgrade, nothing to break. Our work for Anushka Gehi Law and Dan Park Insurance follows that approach. If a framework would add complexity without helping you, we skip it.

Tailwind CSS

A styling approach where design rules live next to the markup. It keeps stylesheets small and consistent across pages, and it makes later changes safer.

Hosting and deploy

Both Vercel and Netlify connect to a GitHub repository. You push a change, the host builds and publishes it, and every deploy has a record and can be rolled back.

  • Vercel for Next.js and Astro sites, one project per site.
  • Netlify for simple client static sites.

Both give you a preview of a change before it goes live, and both handle certificates and a global CDN, so you are not managing servers.

Publish-safety: prod is defined by a commit

A lesson from a law firm site: some pages sat drafted but unapproved for weeks. The old way of publishing uploaded the whole working folder, including files that were never meant to ship. One tired command and unreviewed legal copy is live.

The fix is structural. Production deploys from a git branch, so production is whatever is in that commit. Anything not committed physically cannot go live. For work with a sign-off step, such as legal, medical or regulated marketing, this turns "remember to keep it back" into "do not commit it."

Forms, functions and data

  • Netlify Forms capture inquiries on static sites with no backend to run. Submissions are stored, and we can pass them to your CRM. For text-message consent, the form stores the consent field as proof of opt-in.
  • Serverless functions handle small jobs: sending a lead to GoHighLevel, converting values, generating a PDF. They run only when called, and cost almost nothing at small-business volume.
  • Supabase is our hosted Postgres database when a project needs real data and logins without us standing up infrastructure.
  • Railway hosts a small always-on backend, such as the one behind our AI phone receptionist (see AI phone receptionist).
  • Google Cloud runs services that must stay up around the clock, independent of any laptop.

One caution from a real incident: a free-tier database auto-paused after about a week of inactivity and quietly broke booking for every client using it until someone noticed. If something is business-critical, it should not sit on a plan that sleeps.

Performance: Lighthouse budgets and image weight

We run Lighthouse CI, an automated test that scores speed, accessibility and SEO, and we set budgets so a change that makes a page slower fails before it ships. On one Windows setup we had to test against the built preview server rather than the built-in static server, because the latter returned no paint data. Details like that are why we test the real thing.

Image weight is the most common cause of a slow small-business site, and it is usually invisible. On Dan Park's insurance site, six oversized photo PNGs and a portrait added up to roughly 49 MB, yet they displayed in cards about 380 pixels wide. We resized them to about 800 pixels wide and saved them as WebP. The total dropped to about 574 KB. The hero videos, around 7.5 MB combined, turned out to be fine. We measured before guessing, and the real culprit was the photos.

Our rules of thumb:

  • Size images to the width they display, with room for sharp screens.
  • Use WebP or similar modern formats.
  • Measure which assets actually weigh the most before optimizing.
  • Keep JavaScript small. Static pages with little script score well and rank better.

Mistakes to avoid

  • Choosing a heavy framework for a five-page site.
  • Uploading camera-original images.
  • Deploying by dragging a folder to a host.
  • Putting a business-critical service on a plan that pauses when idle.

When this stack is not the right fit

If you want to edit everything yourself in a visual builder with no developer involved, a hosted site builder may suit you better, and we will say so. If you need a large e-commerce store with inventory, that calls for a platform built for it. For a lead-generating service site, though, simple and fast is hard to beat. Talk to us about what fits.

Ready to bring in more customers?

Tell us about your business and what you need. We'll reply with a plan and a fixed price, usually within one business day.