Deployment Platform Specialist

Comprehensive deployment platform guide covering Vercel (edge-first), Railway (container-first), and Convex (real-time backend).

---

Quick Platform Selection

When to Use Each Platform

Vercel - Edge-First Deployment:

• Next.js applications with SSR/SSG
• Global CDN distribution required
• Sub-50ms edge latency critical
• Preview deployments for team collaboration
• Managed storage needs (KV, Blob, Postgres)

Railway - Container-First Deployment:

• Full-stack containerized applications
• Custom runtime environments
• Multi-service architectures
• Persistent volume storage
• WebSocket/gRPC long-lived connections

Convex - Real-Time Backend:

• Collaborative real-time applications
• Reactive data synchronization
• TypeScript-first backend needs
• Optimistic UI updates
• Document-oriented data models

---

Decision Guide

By Application Type

Web Applications (Frontend + API):

• Next.js → Vercel (optimal integration)
• React/Vue with custom API → Railway (flexible)
• Real-time collaborative → Convex + Vercel

Mobile Backends:

• REST/GraphQL → Railway (stable connections)
• Real-time sync → Convex (reactive queries)
• Edge API → Vercel (global latency)

Full-Stack Monoliths:

• Containerized → Railway (Docker support)
• Serverless → Vercel (Next.js API routes)
• Real-time → Convex (built-in reactivity)

By Infrastructure Needs

Compute Requirements:

• Edge compute → Vercel (30+ edge locations)
• Custom runtimes → Railway (Docker flexibility)
• Serverless TypeScript → Convex (managed runtime)

Storage Requirements:

• Redis/KV → Vercel KV or Railway
• PostgreSQL → Vercel Postgres or Railway
• File storage → Vercel Blob or Railway volumes
• Document DB → Convex (built-in)

Networking Requirements:

• CDN distribution → Vercel (built-in)
• Private networking → Railway (service mesh)
• Real-time WebSocket → Convex (built-in) or Railway

---

Common Deployment Patterns

Pattern 1: Next.js with Database

Stack: Vercel + Vercel Postgres/KV

Setup:

1. Deploy Next.js app to Vercel
2. Provision Vercel Postgres for database
3. Use Vercel KV for session/cache
4. Configure environment variables
5. Enable ISR for dynamic content

Best For: Web apps with standard database needs, e-commerce, content sites

Pattern 2: Containerized Multi-Service

Stack: Railway + Docker

Setup:

1. Create multi-stage Dockerfile
2. Configure railway.toml for services
3. Set up private networking
4. Configure persistent volumes
5. Enable auto-scaling

Best For: Microservices, complex backends, custom tech stacks

Pattern 3: Real-Time Collaborative App

Stack: Convex + Vercel/Railway (frontend)

Setup:

1. Initialize Convex backend
2. Define schema and server functions
3. Deploy frontend to Vercel/Railway
4. Configure Convex provider
5. Implement optimistic updates

Best For: Collaborative tools, live dashboards, chat applications

Pattern 4: Hybrid Edge + Container

Stack: Vercel (frontend/edge) + Railway (backend services)

Setup:

1. Deploy Next.js frontend to Vercel
2. Deploy backend services to Railway
3. Configure CORS and API endpoints
4. Set up edge middleware for routing
5. Use private networking for Railway

Best For: High-performance apps, global distribution with complex backends

Pattern 5: Serverless Full-Stack

Stack: Vercel (frontend + API routes) + Convex (backend)

Setup:

1. Build Next.js app with API routes
2. Initialize Convex for data layer
3. Configure authentication (Clerk/Auth0)
4. Deploy frontend to Vercel
5. Connect Convex client

Best For: Rapid prototyping, startups, real-time web apps

---

Essential Configuration

Vercel Quick Start

vercel.json:

{
  "$schema": "https://openapi.vercel.sh/vercel.json",
  "framework": "nextjs",
  "regions": ["iad1", "sfo1", "fra1"],
  "functions": {
    "app/api/**/*.ts": {
      "memory": 1024,
      "maxDuration": 10
    }
  }
}

Edge Function:

export const runtime = "edge"
export const preferredRegion = ["iad1", "sfo1"]

export async function GET(request: Request) {
  const country = request.geo?.country || "Unknown"
  return Response.json({ country })
}

Railway Quick Start

railway.toml:

[build]
builder = "DOCKERFILE"
dockerfilePath = "Dockerfile"

[deploy]
healthcheckPath = "/health"
healthcheckTimeout = 100
restartPolicyType = "ON_FAILURE"
numReplicas = 2

[deploy.resources]
memory = "2GB"
cpu = "2.0"

Multi-Stage Dockerfile:

# Builder stage
FROM node:20-alpine AS builder
WORKDIR /app
COPY package*.json ./
RUN npm ci
COPY . .
RUN npm run build

# Runner stage
FROM node:20-alpine
WORKDIR /app
ENV NODE_ENV=production
RUN addgroup -g 1001 -S nodejs && adduser -S appuser -u 1001
COPY --from=builder /app/node_modules ./node_modules
COPY --from=builder /app/dist ./dist
USER appuser
EXPOSE 3000
CMD ["node", "dist/main.js"]

Convex Quick Start

convex/schema.ts:

import { defineSchema, defineTable } from "convex/server"
import { v } from "convex/values"

export default defineSchema({
  messages: defineTable({
    text: v.string(),
    userId: v.id("users"),
    timestamp: v.number(),
  })
    .index("by_timestamp", ["timestamp"])
    .searchIndex("search_text", {
      searchField: "text",
      filterFields: ["userId"],
    }),
})

React Integration:

import { useQuery, useMutation } from "convex/react"
import { api } from "../convex/_generated/api"

export function Messages() {
  const messages = useQuery(api.messages.list)
  const sendMessage = useMutation(api.messages.send)

  if (!messages) return <div>Loading...</div>

  return (
    <div>
      {messages.map((msg) => (
        <div key={msg._id}>{msg.text}</div>
      ))}
    </div>
  )
}

---

CI/CD Integration

GitHub Actions - Vercel

name: Deploy to Vercel
on:
  push:
    branches: [main]

jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      - uses: amondnet/vercel-action@v25
        with:
          vercel-token: ${{ secrets.VERCEL_TOKEN }}
          vercel-org-id: ${{ secrets.ORG_ID }}
          vercel-project-id: ${{ secrets.PROJECT_ID }}

GitHub Actions - Railway

name: Deploy to Railway
on:
  push:
    branches: [main]

jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      - run: npm install -g @railway/cli
      - run: railway up --detach
        env:
          RAILWAY_TOKEN: ${{ secrets.RAILWAY_TOKEN }}

GitHub Actions - Convex

name: Deploy to Convex
on:
  push:
    branches: [main]

jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      - uses: actions/setup-node@v3
      - run: npm ci
      - run: npx convex deploy
        env:
          CONVEX_DEPLOY_KEY: ${{ secrets.CONVEX_DEPLOY_KEY }}

---

Advanced Patterns

Blue-Green Deployment (Vercel)

Deploy new version, test on preview URL, then switch production alias using Vercel SDK for zero-downtime releases.

Multi-Region (Railway)

Configure deployment regions in railway.toml:

[deploy.regions]
name = "us-west"
replicas = 2

[[deploy.regions]]
name = "eu-central"
replicas = 1

Optimistic Updates (Convex)

const sendMessage = useMutation(api.messages.send)

const handleSend = (text: string) => {
  sendMessage({ text })
    .then(() => console.log("Sent"))
    .catch(() => console.log("Failed, rolled back"))
}

---

Platform-Specific Details

For detailed platform-specific patterns, configuration options, and advanced use cases, see:

• reference/vercel.md - Edge Functions, ISR, Analytics, Storage
• reference/railway.md - Docker, Multi-Service, Volumes, Scaling
• reference/convex.md - Reactive Queries, Server Functions, File Storage
• reference/comparison.md - Feature Matrix, Pricing, Migration Guides

---

Works Well With

• moai-domain-backend for backend architecture patterns
• moai-domain-frontend for frontend integration
• moai-lang-typescript for TypeScript best practices
• moai-lang-python for Python deployment (Railway)
• moai-platform-auth for authentication integration
• moai-platform-database for database patterns

---

Status: Production Ready Version: 2.0.0 Updated: 2026-02-09 Platforms: Vercel, Railway, Convex

Common Rationalizations

Rationalization	Reality
"I will configure the deployment platform after development is complete"	Deployment configuration affects build output, environment variables, and runtime behavior. Configure early.
"Preview deployments are optional"	Preview deployments catch deployment-specific bugs before production. They cost little and save a lot.
"Environment variables are the same across all environments"	Production, staging, and development need different database URLs, API keys, and feature flags. One set of env vars is a security risk.
"Serverless cold starts are negligible"	Cold starts add 200-2000ms latency on first request. For user-facing APIs, this matters. Measure and mitigate.
"I do not need a rollback strategy, I can just redeploy"	Redeployment takes minutes. Rollback takes seconds. When production is down, seconds matter.

Red Flags

• Production environment variables visible in build logs
• No preview deployment configured for pull requests
• Single environment used for both staging and production
• No rollback mechanism documented or tested
• Build artifacts include development dependencies or source maps

Verification

• Environment variables separated by environment (dev, staging, production)
• Preview deployments configured for pull requests (show platform config)
• Rollback procedure documented and tested (show rollback command or process)
• Production build excludes development dependencies and source maps
• Deployment succeeds from a clean git checkout (no local state dependency)
• Cold start time measured for serverless functions (show timing data)

Refactor Notes

R4 audit verdict (2026-04-23): REFACTOR — shrink triplet to Vercel-only primary; Railway/Convex as documentation-only SPEC: SPEC-V3R2-WF-001 §6.2 line 271 Refactor scope (deferred to future sub-SPEC):

• Elevate Vercel as primary platform with full guidance; demote Railway and Convex to doc-only references
• Move Railway and Convex deep-dives into Level-3 modules
• Add CI/CD pipeline patterns section (GitHub Actions, Vercel CI integration)

This skill is retained in v3.0 but its body will be restructured in a follow-up SPEC.