Stack-Cost

StackCost is the answer to “will this stack actually stay free?”.You pick services,set MAU and data per user, and it returns a per-provider monthly estimate.

Comment

Inspiration

Free-tier dashboards look generous until you ship. Then MAU, egress, and “Hobby is non-commercial” hit at once, and the first real bill (or throttle) is the first time you notice.

Stacksmith already planned what to build. We wanted the next question founders actually ask: what does this stack cost at my scale, and which limit breaks first? StackCost is that check — architecture cost and tier estimation before surprise invoices.

What it does

You pick from a fixed catalog — Supabase, Vercel, Clerk, Neon, OpenRouter, PostHog — then set monthly active users and storage/transfer per user.

POST /api/analyze returns:

  • freeTierExceeded — whether any service needs a paid tier at this load
  • breakdown[] — per provider: cheapest covering tier, estimated monthly cost, and the bottleneck (the limit that decides the tier)
  • totalEstimatedMonthly
  • recommendation — one or two concrete cost-cut lines for this stack and scale

Estimates can be copied as Markdown or shared via a #e= URL fragment (no database). Numbers are a starting point for a pricing-page check, not a quote.

How we built it

Forked Stacksmith for the Build with Stripe Community hackathon (Stripe Dhaka AI Event 2026). Same Next.js 15 + Stripe Projects scaffolding; the planner became an estimator.

  • UI: one client surface (StackCost) — service toggles, MAU/data sliders, result cards, breakdown table
  • AI: OpenRouter with a compact published free-tier reference (lib/tiers.ts) in the system prompt so the model reasons from real allowances instead of inventing limits
  • Reliability: a :free model fallback chain, JSON extraction from messy completions, and sanitization so the breakdown only includes services the user actually picked
  • Infra: OpenRouter + Vercel via the Stripe Projects CLI. No database — every estimate is stateless

Challenges we ran into

  • Credits vs free models: a fresh OpenRouter account can 402 on paid models. We kept a :free fallback and treat slugs as something to re-check on OpenRouter’s model list.
  • Vercel PLAN_REQUIRED: vercel/project needs vercel/hobby first — the CLI tells you, but only after you guess wrong.
  • LLMs and JSON: models wrap JSON in fences or invent extra providers. We extract balanced JSON, validate the shape, and drop rows that aren’t in the selected stack.
  • Pricing that lies by omission: Hobby “free” vs commercial-use clauses; DB size and egress often blow before MAU caps. Encoding those bottlenecks in the prompt mattered more than another UI widget.
  • Cloned .projects/ state: Stacksmith’s Stripe account isn’t ours — re-init against your own account before you can provision.

Accomplishments that we're proud of

  • A demo you can finish in one sitting: pick a stack, set scale, see bill + throttle + a specific bottleneck.
  • Grounding estimates in a real (compact) tier table, not vibes.
  • Shareable, zero-backend estimates (#e= + Markdown report) that still feel like a product.
  • Shipping on free OpenRouter models with retries so the hackathon demo doesn’t die on the first 500/402.
  • Tests for JSON extraction, share encoding, and report rendering — small, but they catch the failure modes that ruin live demos.

What we learned

Provider “free” is a bundle of limits, not a price of $0. The interesting output isn’t a single monthly number — it’s which dimension fails first.

Also: Stripe Projects is a real workflow (plan before deployable, env pull, rotate on leak). LLM apps in a weekend live or die on prompt constraints + parse/sanitize, not on model brand. And stateless by default is a feature when you don’t need accounts.

What's next for Stack-Cost

  • More providers and usage-shaped inputs (events/user, function GB-hours, recordings) instead of MAU + MB only
  • Refreshable pricing from official pages or APIs so TIER_REFERENCE doesn’t rot
  • Side-by-side stack A vs stack B and “what if we drop this service?”
  • Optional saved estimates / team share without giving up the no-login path
  • Clearer disclaimer UX: always confirm against current provider pricing before you commit spend

Built With

Share this project:

Updates

Submission history