Deploying FudFi

FudFi deploys as three independently hosted pieces:

Piece	Host	Project	Root directory
Backend API	Railway	`fudfi` / service `backend`	repo root
App	Vercel	`fudfi-next`	`frontend`
Marketing	Vercel	`fudfi-landing`	`marketing`

Backend on Railway

Railway uses the Railpack builder, which builds the repo root as a monorepo and ignores most of railway.json. So the root package.json MUST have a start script and MUST NOT have a build script (otherwise Railpack tries to build the Next.js frontend for the backend and fails).

Root package.json requirements:

"start": "npm run start:backend" — exists today.
No root build script. Web builds live in frontend/ only.
"engines": { "node": ">=20.9.0" } — Next 16 + Railpack’s default Node 18 otherwise.

Deploy:


railway up --service backend

Verify the new code is live:


curl -X POST https://<backend-host>/api/auth/evm/challenge \
  -H "Content-Type: application/json" -d '{}'
# expect HTTP 400 (validation), not 404

Environment


DATABASE_URL=postgres://...          # Railway Postgres proxy — NO ?sslmode= (db/client.js handles TLS)
ALLOWED_ORIGINS=https://app.fudfi.com,https://fudfi.com,https://www.fudfi.com,http://localhost:3001
NODE_ENV=production
GEMINI_API_KEY=...
HF_API_KEY=...
TWITTER_CLIENT_ID=... TWITTER_CLIENT_SECRET=... TWITTER_CALLBACK_URL=...
CLOUDINARY_CLOUD_NAME=... CLOUDINARY_API_KEY=... CLOUDINARY_API_SECRET=...
SOLANA_RPC_URL=https://api.mainnet-beta.solana.com

Redis

The backend supports Redis via REDIS_URL and falls back to an in-memory cache. Production uses a Railway service named Redis, wired on backend as:


REDIS_URL=${{Redis.REDIS_URL}}

Omit REDIS_URL locally; in-memory is fine.

App on Vercel

Project fudfi-next (team team_VOLeewSIrAYhkcVM9HBePF3N). Its Root Directory is baked to frontend in project settings.
Deploy from the repo root (vercel --prod), never from inside frontend/.
Commits must be authored fortunesayspump <fortuneze0@gmail.com>. Any other author gets the deployment blocked (seatBlock TEAM_ACCESS_REQUIRED).
The first vercel --prod attempt often says “Not authorized” — just retry; it passes.
Custom domain: vercel alias set <deployment-url> app.fudfi.com.
SSO protection must stay deploymentType: preview; anything stricter walls app.fudfi.com behind a Vercel login page.

Required env on the project:


NEXT_PUBLIC_API_URL=https://backend-production-03de6.up.railway.app/
NEXT_PUBLIC_REOWN_PROJECT_ID=...
NEXT_PUBLIC_EVM_CHAIN=testnet   # or mainnet

Docs on Vercel

docs/ is a standalone Nextra app with its own package.json and package-lock.json. It is not part of the root npm workspaces. To host it, create a separate Vercel project with Root Directory = docs (install and build run inside docs/).


cd docs && npm install && npm run build   # local verification

Native binaries (the recurring deploy killer)

This repo is developed on Windows but deployed on Linux. npm writes the lockfile for the local platform, so Linux-only native modules go missing remotely. If a deploy fails with Cannot find module '../*.node' or Could not load the "sharp" module, pin the Linux packages in optionalDependencies and regenerate the root lockfile:

frontend: lightningcss-linux-x64-gnu, @tailwindcss/oxide-linux-x64-gnu (Tailwind v4), plus @x402/core, @x402/evm, @x402/svm.
backend: @img/sharp-linux-x64 (pinned to sharp’s exact version).

After any dependency change:


npm install                       # at repo root
node -e "Object.keys(require('./package-lock.json').packages).filter(p=>/linux-x64/.test(p)).forEach(console.log)"
# commit BOTH package.json AND package-lock.json

There is intentionally no frontend/package-lock.json. Do not recreate it — it was stale and confused Next/Vercel. The root lockfile is the only lockfile.