Should you build your own Zapier?

Build me a personal automation runner to replace Zapier, for the handful of pipelines I actually use. Requirements:

- One Node process on my VPS: Express for incoming webhooks, node-cron for
  schedules. A pipeline is a plain JS file in pipelines/ exporting a trigger (cron
  expression or webhook path) and an async run(payload).
- Steps are just code: fetch() calls to the APIs I need, every key in .env. Adding
  an integration means writing 20 lines, not finding a connector.
- Every run logged to SQLite (better-sqlite3): pipeline, trigger payload, status,
  duration, error. Failures retry 3 times with exponential backoff via p-retry.
- A /runs page (bound to localhost) showing the last 100 runs with status and an
  expandable payload/error view, plus a re-run button.
- After the final failed retry, notify me via an ntfy topic in .env.
- Webhook URLs include a per-pipeline secret token; reject anything else.
- Ship two example pipelines: a cron one (daily 08:00, fetch a weather API, push
  to ntfy) and a webhook one (POST in, append a row to a CSV).
- No accounts, no telemetry, no visual editor.
- Out of scope: OAuth flows for third-party apps (token-based APIs only) and any
  connector catalog. A service that only offers OAuth will cost an afternoon each,
  say so in the README.
- README: systemd unit, how to add a pipeline, .env layout.