From Replit to its own infrastructure, with an 85% smaller bill
The Wellu corporate platform was born on Replit, built with AI, and the monthly cost kept growing. We moved the product to its own infrastructure, without rewriting from scratch and without taking clients offline.
- Client
- Wellu
- Sector
- Health
- Line
- Product engineering
- Period
- Mar. to Apr. 2026 (migration); ongoing evolution to date
- Deliverable
- AI-built app · migration
Technology
- React
- TypeScript
- Vite
- Node.js and Express
- PostgreSQL with per-company isolation
- Drizzle ORM
- Docker and Caddy
- AWS S3
- GitHub Actions
A product that worked, locked into an ever more expensive platform
The platform for companies was validated the fastest way possible: built with the Replit AI agent. It worked. In six months it gained leaderboards, a feed, billing and reports, and started serving real companies.
But hosting, database, files and development were locked into the platform, with the bill going up every month. The code, written over thousands of agent runs, had login tied to Replit and secrets stored only in the tool's dashboard.
A phased migration, without rewriting from scratch
- Dependency assessmentA map of everything tying the product to Replit, ranked by risk, with a rollback plan for each phase.
- Own loginEmail and password with server-side sessions, without a single user losing access in the switch.
- Files off the platformOwn storage, with temporary links for private files and image compression.
- A portable productContainerized application, deployed from GitHub to staging and production, with no dependency on a single tool.
- Data isolated per companyA review of client isolation in the database and secrets moved out of the tool's dashboard.
- Evolution with processCode review and separate environments. Since then we keep shipping improvements.
Built with AI, taken to production with engineering
The first version was born with an AI coding agent. Our role was to turn that product, which already had clients, into software the company controls: predictable cost, reviewed security and evolution without depending on a single tool.
We also used AI to read and map a large codebase written by an agent, but every change went through human review.
The same product, with an 85% smaller infrastructure bill
The same product, in Wellu's hands, with a monthly infrastructure cost about 85% lower and client companies using it normally throughout the switch.
Has your AI-built app become expensive?
Book 45 minutes on Teams with a specialist. We understand your context before proposing anything.
Next case: Itaú