Skip to content

HOSTING.md

Latest commit

 

History

History
213 lines (159 loc) · 7.74 KB

HOSTING.md

File metadata and controls

213 lines (159 loc) · 7.74 KB
Raw

Hosting

Pick one provider — all-in-one. The app is a single Docker container with one persistent volume (/data for SQLite). No Redis, no separate database, no queues required until you cross ~50 paying users.

Hetzner Cloud — best price/performance

Recommended for 50-1000 users. ~$28/mo for a CCX23 (4 vCPU dedicated, 16 GB RAM, 160 GB SSD, 20 TB egress).

  1. Create a Hetzner account → Project → Servers → "Add Server".
  2. Choose image: Docker CE on Ubuntu 22.04. Type: CCX23 (or CX22 for smaller scale). Add SSH key.
  3. SSH in: ssh root@your-server-ip
  4. git clone https://github.com/IamRamgarhia/SEO-Tool.git
cd seo
docker compose up -d
  5. Point your domain's A record at the server IP.
  6. Add Caddy or Nginx in front for SSL — easiest: 
    docker run -d --name caddy \
  -p 80:80 -p 443:443 \
  -v caddy_data:/data \
  -e DOMAIN=seo.yourdomain.com \
  caddy caddy reverse-proxy --from seo.yourdomain.com --to localhost:3000

Backup: docker run --rm -v seo_seo-data:/data -v $(pwd):/backup ubuntu tar czf /backup/seo-backup.tar.gz /data

Railway — easiest managed deploy

Recommended for 0-200 users when you want zero ops.

  1. Push the repo to GitHub.
  2. railway.app → New Project → Deploy from GitHub repo.
  3. Railway auto-detects the Dockerfile. Add a persistent Volume mounted at /data.
  4. Set env vars under Settings:
    • SEO_DB_PATH=/data/data.db
    • APP_PASSWORD=<your-password> (optional)
    • Any OAuth / AI keys you want pre-baked
  5. Add a custom domain (Settings → Domains).

Cost: ~$5/mo Hobby plan ≈ 512 MB RAM (tight, fine for solo use). $20+ Pro for real traffic.

Hostinger VPS — cheapest

Recommended for tinkerers comfortable with manual ops.

  1. Hostinger → VPS Hosting → KVM 2 (2 vCPU, 8 GB) for $7/mo or KVM 4 (4 vCPU, 16 GB) for ~$15/mo on the 24-month plan.
  2. Choose OS template: Ubuntu 22 with Docker.
  3. SSH in, then same as Hetzner steps 4-6 above.

Catch: shared CPU on the cheapest tiers (KVM 1) means Playwright workloads can stall under contention. KVM 2 or higher.

DigitalOcean — middle ground

  1. New Droplet → Marketplace → Docker on Ubuntu 22.04.
  2. Size: 8 GB RAM / 4 vCPU ($48/mo) is the sweet spot.
  3. Same Docker compose steps as Hetzner.

Or use DigitalOcean App Platform with the Dockerfile and a connected managed Postgres database — easier but more expensive (~$75/mo).

Self-hosted on your own computer — £0, and the most private

Use the README's Docker option. The tool runs entirely on localhost:3000. No data leaves your machine unless you explicitly connect Google APIs or an external AI provider.

Costs: nothing. The catch is that it's only reachable from that machine, and only while it's switched on.

Reaching it from your phone, still free

Cloudflare Tunnel is the simplest way, needs no open ports on your router, and gives you HTTPS:

# One-time: install cloudflared, then
cloudflared tunnel --url http://localhost:3000

That prints a public *.trycloudflare.com URL. For something permanent, create a named tunnel and point a domain at it — still free.

Tailscale is the alternative if you'd rather it stay private: your devices see the tool, the public internet never does.

Set APP_PASSWORD before you expose it either way. A tunnel makes your install reachable by anyone who learns the URL, and without a password they can use it.

Good for: personal SEO, one or two client sites, anyone who'd rather own their data outright. The limitation is real though — scheduled audits, rank checks and the daily agent only run while your computer is awake. If you want the automation to actually be automatic, you need something that stays on.

Genuinely free, always on

Oracle Cloud Always Free is the only major cloud with a free tier that fits this tool: up to 4 ARM cores and 24 GB RAM, a persistent disk, and no expiry. That is far more than this needs — a 1 GB VPS runs it fine.

Install Docker on the instance and follow the Hetzner steps above; they are identical from that point. Open port 443, put Cloudflare in front, set APP_PASSWORD.

Two honest caveats: ARM capacity in popular regions is often unavailable, so you may have to retry or pick a quieter region; and Oracle's signup asks for a card for identity even on the free tier.

Railway gives $5 of monthly credit, which this tool will roughly consume — expect a small bill rather than free. Render runs Docker with a real free tier, but persistent disks are a paid add-on, and without one your database is wiped on every deploy. Fly.io no longer has a free tier as of 2026 — new accounts get a short trial, then roughly $2-5/month.

Vercel + Supabase / Neon — why this doesn't work

Worth stating plainly, because it's the obvious thing to try and it would waste an afternoon.

Vercel can't run this. Not a configuration problem — four separate blockers:

  1. The database is a file. better-sqlite3 opens data.db in WAL mode on local disk. Vercel's filesystem is ephemeral and each request may hit a different container, so your data would vanish between requests.
  2. Playwright. Rank checks, SERP scraping and GBP scraping drive a real headless Chromium. It doesn't fit in a serverless function.
  3. The scheduler needs a process that stays alive. Audits, rank checks, monitoring and the daily agent are started from instrumentation.ts at boot. Serverless has no always-on process, so nothing would ever run on a schedule — which is most of the point.
  4. output: "standalone" builds a self-contained server for a container, not Vercel's runtime.

Supabase and Neon are PostgreSQL, and this app is SQLite — see "When to migrate to PostgreSQL" below for what that actually costs. They are excellent databases; they're just not a drop-in here.

If you want it hosted and free, use Oracle Cloud Always Free. If you want it hosted and effortless, a $5 VPS is the shortest path.

After deploy — production hardening

  1. Cloudflare in front (free) — point your domain through Cloudflare. Saves 60-80% of bandwidth + adds DDoS protection + free SSL.
  2. Backup the SQLite file — daily cron: 
    sqlite3 /data/data.db ".backup /backups/$(date +%F).db"
  3. Set APP_PASSWORD in the env so the UI is gated. Without it, anyone who finds the URL can use the tool.
  4. Cap browser concurrency in Settings → Browser if you hit RAM pressure. Default 2 fits in ~1 GB; raise to 4 on a 4+ GB machine.
  5. Enable Playwright proxy rotation if you scrape SERPs at volume — Settings → Browser → paste proxy list.

When to migrate to PostgreSQL

Stay on SQLite while:

  • You're the only user, OR
  • You have <50 active users with <30 concurrent requests at peak.

Consider PostgreSQL when:

  • You see SQLITE_BUSY errors in production logs.
  • Daily-agent runs take >30 minutes (writes are queuing).
  • You want to run 2+ app instances behind a load balancer.

It is not a config swap. This page used to say "a DATABASE_URL env var + the drizzle config swap — about 1 day of work". That was wrong, and measured rather than guessed:

  • 72 tables are declared with sqliteTable from drizzle-orm/sqlite-core
  • 42 of 62 migrations use SQLite-only SQL — AUTOINCREMENT, unixepoch(), PRAGMA
  • src/db/client.ts opens a better-sqlite3 handle and sets WAL mode
  • Drizzle's SQLite and Postgres builders are different modules with different types, so the schema file is a rewrite, not an edit

Drizzle does support both, and the shape of the schema carries over. But this is a port measured in weeks, not a day. Nobody has done it, so treat any estimate — including this one — as an estimate.