My Indie Stack in 2026, and Why
I build and run seven projects on my own: three products (ReadSprint, InvoiceAgent and Feedbackly) and four games (Ash & Oath, Motive, Piko and Venture Life), plus this site.
When you're one person, the stack isn't a technical preference. It's how much you can maintain at once. Every service is something else that can break at 2am, and every unfamiliar tool is a context switch. So the rule I've ended up with is the same few tools everywhere, unless a project has a real reason to differ. Here's what that looks like.
At a glance
| Layer | Products | Games |
|---|---|---|
| Framework | Next.js | Vite, with Three.js or Phaser, or React |
| Language | TypeScript | JavaScript (Motive is TypeScript) |
| Database | Postgres: Prisma, or Supabase for ReadSprint | Neon Postgres, with Drizzle in three of the four |
| Auth | Clerk (Supabase Auth for ReadSprint) | Clerk, optional |
| Resend | ||
| Payments | Dodo Payments | Dodo Payments |
| Styling | Tailwind CSS | Hand-written CSS |
| Hosting | Vercel | Vercel, plus Vercel Blob for downloads |
| Desktop | Electron |
Products: Next.js
All three products are Next.js apps. For a product, I want the marketing site, the app, the API and SEO pages in one codebase with one deploy. Route Handlers cover the API (tutorial), and server rendering means each product's marketing pages are crawlable without extra work.
If you build with AI coding agents, Next.js is the perfect place to start. Its feature set is a big part of that, including support for the latest React releases.
Games: Vite, mostly without an engine
The games skip Next.js entirely. They're mostly a single, large client-side app with a few API endpoints, so Vite's fast dev server and simple builds fit better. On top of Vite:
- Ash & Oath uses Three.js directly for its 2.5D world (how).
- Venture Life uses Phaser 3 for its simulation and game world.
- Motive is React and TypeScript, with React Three Fiber for its 3D scenes.
- Piko is plain JavaScript and the DOM.
Motive is in the middle of a bigger change. Its cases started out as 2D scenes, and I'm converting all of them to 3D. The rooms are modelled in Blender, and I'm bringing Unity in alongside it to render them. It's still a work in progress.
Data: Postgres everywhere
Every project stores its data in Postgres, accessed three different ways, which is a good illustration of picking tools per project:
- Prisma for InvoiceAgent and Feedbackly. Its schema file is a readable description of the whole product, and its migrations are dependable. Both products have relational data where constraints matter: invoices, payments and reminders, or posts, votes and followers.
- Neon for the games, with Drizzle in Motive, Piko and Venture Life (Ash & Oath queries Neon directly). Game saves are simpler, Drizzle is lighter, and Neon's serverless driver suits Vercel functions.
- Supabase for ReadSprint, which also uses Supabase Auth and Storage (for book covers and audio), so the database, auth and files come from one service.
If I started ReadSprint today, I wouldn't choose Supabase. I'd use the same stack as the other products, Prisma and Clerk, because it's easier to set up and use.
Auth: Clerk
Clerk handles sign-in for InvoiceAgent, Feedbackly and all four games. Auth is the clearest example of something I never want to build myself: the sign-in screens, social logins, sessions and account management are all included.
In the games, accounts are optional. You can play anonymously with local saves, and signing in adds cloud saves, a place on the leaderboards and the option to buy add-ons. Making accounts optional was more work, but forcing sign-up before someone has even tried a game would cost far more players.
Email: Resend
Resend sends InvoiceAgent's invoices and reminders, Feedbackly's notifications and ReadSprint's newsletter. It's a simple API with good deliverability, and you can design emails as React components.
Payments: Dodo Payments
Every product and game takes payments through Dodo Payments, whether subscriptions for the products or one-time unlocks in the games. As a merchant of record, Dodo handles VAT and sales tax worldwide, which is the only reason a one-person business can sell globally without a tax department. I wrote more in pricing a solo-built SaaS.
Hosting: Vercel
Everything deploys to Vercel: push to main and it's live. Scheduled jobs (invoice sending, reminders, newsletters, billing reconciliation) run as Vercel cron jobs calling API routes, and Vercel Blob hosts the desktop game installers.
Desktop: Electron
Piko, Ash & Oath, Motive and Venture Life all have Electron builds for macOS and Windows. Electron isn't small, but it lets the exact same web code ship as a desktop app (how).
AI tools in the pipeline
AI shows up in three places: inside a product, in making the games, and in writing the code.
- In products: ReadSprint generates its summaries and quizzes with OpenAI models through the Vercel AI SDK (how).
- In production: game music and sound effects are generated with ElevenLabs, and intro videos and some art with fal.ai.
- In development: AI coding agents are part of how I build everything. That deserves its own post: how I build with AI coding agents as a solo developer.
What I'd change
The database is where my choices are settling down. For anything new, my go-to is now Prisma as the ORM with Neon as the database.
Pricing plays a role. InvoiceAgent and Feedbackly are hosted on Prisma Postgres, which charges per operation. Neon charges for compute, based on CPU usage. Combining them keeps the part of Prisma I like, the schema and migrations, on a database billed the way I prefer.
The real rule
The specific tools matter less than the principle: choose boring, managed services for everything that isn't your product. Auth, email, payments, hosting and the database are solved problems. Every hour I don't spend on them goes into the things people actually use.