Self-hosting

Pylon runs entirely on Cloudflare: Workers (app + probes), D1 (SQLite database), R2 (logo uploads) and Cron Triggers (a once-a-minute scheduler that runs every check). There are no Durable Objects, so nothing is billed per monitor or per open page.

Prerequisites

• Node.js 22+ and pnpm
• A Cloudflare account (npx wrangler login)

Choose a plan

Pylon runs on either Cloudflare Workers plan with the same code. Pick the column that matches your account:

Deploy

Option A — guided (recommended)

git clone https://github.com/jackh54/Pylon && cd Pylon
pnpm install
npx wrangler login
pnpm setup

pnpm setup asks paid or free, then creates the D1 database, the R2 bucket and (Paid) the alert queue, writes the ids into wrangler.jsonc, applies migrations, builds, deploys and sets SESSION_SECRET. Non-interactive:

pnpm setup --plan paid --url https://status.example.gg --yes
pnpm setup --plan free --url https://status.example.gg --yes
pnpm setup --dry-run        # show what would happen

Option B — manual, Workers Paid

npx wrangler d1 create pylon            # paste the id into wrangler.jsonc → d1_databases[0].database_id
npx wrangler r2 bucket create pylon-uploads
npx wrangler queues create pylon-notify
pnpm db:migrate                         # applies drizzle/*.sql to the remote D1
# set "APP_URL" in wrangler.jsonc to your public URL
pnpm deploy
npx wrangler secret put SESSION_SECRET  # openssl rand -hex 32

Option C — manual, Workers Free

Same as above, minus the queue, plus two edits in wrangler.jsonc:

1. Delete the block between // BEGIN paid-only and // END paid-only (the queues section). Alerts are then sent inline.
2. Set "PBKDF2_ITERATIONS": "100000" — the Free plan rejects higher iteration counts.

npx wrangler d1 create pylon
npx wrangler r2 bucket create pylon-uploads
pnpm db:migrate
pnpm deploy
npx wrangler secret put SESSION_SECRET

Instance settings

After the first deploy

• Open the URL Wrangler printed, register the first account (it becomes the owner of the first team).
• Put that URL into APP_URL if you left it empty, and redeploy.
• Optional secrets: STEAM_API_KEY (edge A2S checks), RESEND_API_KEY + EMAIL_FROM (email alerts and subscriber emails), DISCORD_CLIENT_ID + DISCORD_CLIENT_SECRET (Discord login).
• Make the instance invite-only by setting ALLOW_SIGNUP to "false".

Switching plans later

Upgraded to Paid? pnpm setup --plan paid adds the queue and raises the hashing cost (existing passwords keep working; they are re-hashed on next change). Downgrading: pnpm setup --plan free. Redeploy afterwards.

Local development

cp .dev.vars.example .dev.vars        # set SESSION_SECRET
pnpm db:migrate:local
pnpm dev                              # http://localhost:5173 – runs inside workerd with local D1/DO/R2

Local dev does not run cron schedules. Run one scheduler minute (checks every due monitor for ~50 s): curl "http://localhost:5173/cdn-cgi/handler/scheduled?cron=*+*+*+*+*".

Architecture

Browser ──HTTP──▶ Worker (React Router SSR + API)
                     │  ├─ D1: users, orgs, monitors, heartbeats (30 d), pages, incidents, daily_stats
                     │  ├─ R2: uploads
                     │  └─ Cache API: public pages and status.json (20 s), polled by open pages
                     └─ Cron every minute: claims due monitors, runs edge probes (6 at a time),
                            fires watchdogs for push/relay monitors that went quiet
Relay (Node) ──pull config / push results──▶ Worker

Plans, limits & costs

Workers Paid (recommended, $5/month): 50M D1 writes/month included (~350 monitors at 60 s), Queues for retried alert delivery, 10 MB scripts, 30 s CPU per request, higher PBKDF2 cost (PBKDF2_ITERATIONS defaults to 600000 in wrangler.jsonc). This is the configuration the repository ships with.

Workers Free: works too. Remove the queues block from wrangler.jsonc (alerts are sent inline) and stay under 100k D1 writes/day (~20 monitors at 60 s) and about 45 edge checks a minute (each cron invocation may make 50 outbound requests). Password hashing is the same on both plans: Workers limits a single PBKDF2 call to 100000 iterations, so Pylon chains blocks to reach PBKDF2_ITERATIONS.

• Each check writes three rows (the scheduling claim, the snapshot on monitors, the raw heartbeats row); a daily rollup is flushed every 5 minutes.
• Raw history is retained 30 days per monitor; rollups 400 days.
• Sub-minute intervals work because each cron invocation stays alive for ~50 s and looks for due checks every 5 s.
• The edge cannot send UDP or reach private IPs — use a relay.
• Custom hostnames for many tenants: Cloudflare for SaaS (100 hostnames free, then per-hostname pricing).

Custom domains for status pages

Three options, from simplest to most flexible: Worker Custom Domains for hostnames on your own Cloudflare account, the free edge proxy in deploy/edge (Caddy on any VPS, for customer-owned domains), or Cloudflare for SaaS (100 hostnames included). See [Status pages](/docs/status-pages) for details.

Updating

git pull && pnpm install && pnpm db:migrate && pnpm deploy