By Flavio Copes

I built stackplan.dev, a deployment stack advisor. Describe your app, get a recommended hosting stack with real cost estimates. Why I built it and how.

~~~

I shipped a new product: StackPlan.

You describe your app — through a questionnaire, a free-text prompt, or by pointing it at a GitHub repo — and it recommends a deployment stack: where to host it, what services to use, and what it will cost. Not just today: it forecasts costs at 100, 1k, 10k, and 100k daily active users.

I mentioned it briefly in my Summer of Code recap. This post is the full story.

Why I built it

Coding is increasingly solved by AI agents. Architecture and deployment are not.

An agent can build you a full app in an afternoon. Then you ask “where should I deploy this?” and you get a generic answer, with no real pricing behind it. Hosting costs are scattered across dozens of pricing pages, free tiers change constantly, and the difference between “$0/month” and “$400/month surprise bill” is often one architectural decision made on day one.

I wanted a tool that answers three questions with actual numbers: where should this app run, what will it cost, and what happens when it grows.

What it does

The core flow: describe your app at stackplan.dev/new, get a report.

The report centerpiece is an interactive stack canvas, Railway-style: your services as draggable nodes, connected with edges, each with a tier knob. Change a tier and the cost recomputes live in the browser.

My favorite part is the engineering dial: a five-stop control that morphs your stack from “Duct Tape” to “Overkill”. Same app, five philosophies, five price tags. It makes the trade-off visible in a way a pricing table never could.

Under the recommendations sits a curated knowledge base of 16 providers — Cloudflare, Vercel, Railway, Fly.io, Hetzner, AWS, Netlify, Render, and more — with their services and plan tiers. The cost math is deterministic, computed from that data. The LLM only writes the rationale on top: it never invents prices.

There’s more around the edges: provider comparisons, case studies of real indie businesses and their stacks, free deployment tools, and an agent API — the same recommendation engine as JSON, so your coding agent can ask “where should I deploy this?” mid-session.

How I built it

StackPlan dogfoods the kind of stack it recommends.

It runs on Cloudflare Workers, with the AHA stack: Astro SSR for rendering, htmx for dynamic updates, Alpine.js for client-side interactivity. No React, no build-heavy frontend.

The data layer is Cloudflare D1 — the provider knowledge base, users, reports, everything lives in a SQLite database at the edge, accessed through Drizzle ORM. KV caches LLM responses. Every public form is protected by Turnstile. LLM calls go through Cloudflare AI Gateway, so I get logging and caching on every request.

Auth is Better Auth, billing is Polar, transactional email is Resend. Deploys are one command with wrangler.

And the build itself: from empty folder to production in 6 days, working with Cursor agents. The agents did the heavy lifting — planning, implementing, testing. My job was direction and review. There’s even an AI agent in production now: it checks provider prices daily and files update proposals when something changes.

What you can buy

StackPlan is free to explore. Every report shows your top recommended stack in full — canvas, cost estimate, scaling forecast — with the alternative stacks blurred.

Two ways to unlock more:

• Report Unlock, $20 one-time — every alternative stack for one report, full scaling forecasts, yours forever. No subscription.
• Pro, $20/month — every report fully unlocked, up to 10 saved apps, cost re-evaluation when providers change pricing, and infra update alerts for your stacks.

If you’re about to deploy something and you’re not sure where, give it a try. And if you have feedback, I’d love to hear it.

~~~

Related posts about news: