Deployment

How Hatch is deployed in production: Vercel (web) + Railway (api) + Supabase (Postgres) + BNB Chain (contracts).

Also read: Vercel deploy runbook for the hands-on playbook.

Platform map

Web — Vercel

Production URL

• gohatch.fun (once domain is connected)
• Fallback: hatch-*.vercel.app

Env vars (Vercel dashboard → Project Settings → Environment Variables)

Deploy Protection

Currently SSO-gated. Disable SSO for the public site to return 200 instead of 401. Settings → Deployment Protection → disable.

Preview URLs

Every PR gets a hatch-<slug>-<hash>.vercel.app URL. Preview deployments use the Preview env; set HATCH_API_BASE_URL to staging if you want previews isolated.

Promote to production

Vercel dashboard → pick a green deployment → "Promote to production". No code change needed. Rollback is the same UI flipping to an older green.

Build command

pnpm --filter @hatch/web build

Uses Turborepo — the web build depends on @hatch/ui, @hatch/shared, and implicitly @hatch/sdk (types only).

API — Railway

Production URL

• api.gohatch.fun (once DNS is set)

Env vars

Deploy flow

1. Push to main.
2. Railway detects the push, runs pnpm install && pnpm --filter @hatch/api build.
3. Starts the API with pnpm --filter @hatch/api start.
4. Health check /health must 200 within 30s or deploy is aborted.

Rollback

Railway dashboard → previous deployment → "Redeploy". Takes ~90 seconds.

Scaling

• Vertical: upgrade plan in Railway dashboard.
• Horizontal: enable multiple replicas; ensure DB connection pool (pg.Pool) doesn't saturate Supabase's connection limit.

Postgres — Supabase

Pool URLs

Use the pooler (pgBouncer) URL, not the direct URL:

postgres://postgres:<pw>@db.<project>.supabase.co:6543/postgres

Pooler handles the connection count spikes during traffic bursts.

Migrations

Deployed ahead of app code:

# Locally, before merging the PR:
pnpm --filter @hatch/api db:generate    # reviews SQL diff
git commit -m "db(…): new migration …"

# In production, after the PR merges:
pnpm --filter @hatch/api db:migrate     # one-off from Railway console

See DEV-GUIDE.md — Database for the migration rules.

Backups

Supabase's daily backups (Pro plan). Retention: 7 days. Manual snapshot before destructive migrations.

Contracts — BNB Chain

Testnet (today)

cd packages/contracts
forge script script/DeployAll.s.sol \
  --rpc-url $BSC_TESTNET_RPC_URL \
  --private-key $DEPLOYER_PRIVATE_KEY \
  --broadcast --verify \
  --etherscan-api-key $BSCSCAN_API_KEY

Mainnet (gated on Sprint C.7)

Same script, different RPC. Requires:

• ✅ C.6 audit complete (OZ / Hacken / ToB)
• ✅ I.5 legal opinion
• ✅ Multisig deployer (not EOA)
• ✅ DeployAll dry-run simulated on testnet with production parameters

Post-deploy:

1. HatchRegistry.setAttester(<hot wallet>).
2. Rotate attester key immediately if deploy log touched it.
3. Update HATCH_ATTEST_ADDRESS on Railway.

DNS + CDN — Cloudflare (planned)

• gohatch.fun → Vercel (proxied or CNAME).
• api.gohatch.fun → Railway.
• Cloudflare WAF: rate-limit beyond API limits; block obvious scraper UAs.
• Redirect www.gohatch.fun → gohatch.fun.

Canary + rollback

Web

1. Deploy via Vercel. First ~5 mins go to ~10% of traffic via Vercel's built-in canary on the production domain. No manual steps.
2. Watch Sentry errors + /api/health from a preview URL.
3. If regression detected → Vercel dashboard → promote previous green.

API

1. Railway doesn't canary natively. Instead, deploy to a staging Railway service first.
2. Run smoke tests against staging URL.
3. Promote to prod by switching the HATCH_API_BASE_URL env on Vercel.
4. Rollback = flip HATCH_API_BASE_URL back to the previous service URL.

Monitoring

• Uptime: UptimeRobot or Better Stack pinging /health every 60s.
• Errors: Sentry on both web + api (planned — L.6).
• Logs: pino JSON → Railway logs (+ optional Logtail/Grafana Loki).
• Metrics: OpenTelemetry OTLP → (TBD — Grafana Cloud or New Relic).
• Status page: /admin/status (internal) + /transparency (public).

Cost envelope (target)

Before any revenue (0.5% LP fee post-graduation). Operating on personal funds until mainnet.