Routiq Labs · the stacks
Every app does the same nine jobs. What changes is whether you rent the tool, own it, or get forced off it. Here are all three answers side by side, in the same order as the reels, and the honest note on each one about why it moves.
If you commented STACKS, this is the page: the full answer on when a self hosted stack is worth it and when it quietly costs you more than it saves. Start with Part I.
Part I. The short answer
If any single line on the right is true, stop there. It does not matter how many on the left are also true.
Self host the ones where the bill hurts and the blast radius is small. Rent anything that touches money, deliverability, or a pager you do not carry.
Yes, if
No, if
Part II. Before you touch anything
Three answers, and you almost certainly want the first one. Read the catch before the cost.
Part 1
When: You have shipped something and it has users you can count.
Mostly free tiers
The catch: Nothing. Stay here far longer than you think. Every swap below is something you do because you were forced to, not because it is better.
Part 2
When: A limit or an invoice forced your hand. Not before.
Whatever the invoice says
The catch: Six of the nine move and three do not. Migrating early costs you the thing you were trying to save.
Part 3
When: One bill has started to sting, or the data cannot leave your box.
About $15 a month
The catch: You just became your own DevOps team. Nine services are yours to patch, back up and restore at 2am. Self host one or two, not nine.
Part III. The table
The job numbers never change. Only the answer under them does. Where a cell says stays, that is the answer, not a gap.
It's open source, so owning it is a Docker command. Four people told me Neon at scale and they were right.
Open models are free, private and close enough now. At scale nothing beats it, whatever the comments say about Codex.
The one swap everybody agrees on. Seven people asked for Cloudflare by name.
Self hosted, two subscriptions collapse into one server. At scale you don't swap it for Render, you swap it for a bill.
Nobody self hosts card processing. You'll IPO before you outgrow it.
Kokoro is 82M params, Apache 2.0, runs on a CPU. It's also the only tool nobody argued with.
You can self host it. You probably won't. At volume Telnyx is the same job for roughly half.
GlitchTip is Sentry compatible, so you just swap the DSN. At scale one dashboard eats the whole monitoring stack.
PostHog themselves say the self hosting maths rarely works. Believe them.
Three rows never move: Stripe, and at scale Claude and ElevenLabs. Nobody self hosts card processing, and the two tools people argue hardest about are the two nobody actually leaves.
Row 01 is the one everybody argues about, so it has a page of its own: Supabase or Neon, and why it’s usually the wrong question →
Part IV. Going deeper
Every claim above has a source. These are the ones worth your time before you move anything.
The official Docker compose stack. This is the whole of row 01.
Rows 03 and 04 collapsed into one box. Preview deploys included.
Row 02. Pull an open coding model and point your editor at localhost.
Row 06. 82M params, Apache 2.0, runs on a CPU.
Row 08. Sentry compatible, so the migration is swapping a DSN.
Row 09. PostHog themselves publish why self hosting rarely pays.
The other half
Picking the right stack will not save an app with secrets in the repo, a 2,500-line file and one commit called “updates”. That is the other page: the five mistakes, the two repos side by side, and the six stages.
Read: build it properly →Not sure which column you are in?
A skill that places you in a column, names the swaps with a real forcing function behind them, and talks you out of the rest. Most people should be in column one and are being argued out of it by comment sections.
mkdir -p ~/.claude/skills/pick-a-stack && curl -fsSL https://labs.routiq.ai/skills/pick-a-stack/SKILL.md -o ~/.claude/skills/pick-a-stack/SKILL.md
what stack should I use?
Get the next one
Every stack gets the same treatment: what it costs at ten users, what it costs at ten thousand, and the one that forces the swap. One email per drop, nothing between.