Can AI replace Mara?

Build me a self-hosted lifecycle email agent to replace Mara, for one SaaS product. Requirements:

- Node + Express + better-sqlite3 on my VPS, one SQLite file holding contacts,
  events, sequences, drafts, sends, and suppressions. node-cron ticks once a minute,
  no queue service.
- POST /events takes {type, email, properties} with a bearer token from .env. Every
  event upserts a contact and appends a row I can read back later.
- Sequences are JSON files in ./sequences/: an entry event type, then steps with a
  wait ("2d"), an intent in plain words ("nudge them to connect their first
  integration"), and an optional skip-if event type. Ship two: welcome and
  trial-ending.
- Each tick, find contacts due for a step, drop anyone whose skip-if event already
  fired, then draft with the Anthropic API (key in .env): pass ./voice.md plus that
  contact's last 20 events, get back a subject and a plain-text body.
- Drafts land as pending. A localhost-only admin page lists them with subject, body,
  recipient, and the step that produced it, with Approve and Reject buttons. Nothing
  sends without a click.
- Approved mail goes out through Resend or SMTP from .env, one send row per recipient
  with the provider message id. Bounce and complaint webhooks write to a suppression
  table checked before every send. One-click unsubscribe link in every footer, no
  login.
- Cap at 2 emails per contact per 7 days across all sequences, enforced in code
  before drafting.
- Out of scope: A/B variants, reply parsing, open tracking. Do not build multi-user
  accounts or a visual sequence builder, this runs for one product and one operator.
- README with the deliverability homework: SPF, DKIM, and DMARC on a subdomain like
  mail.example.com, a two-week warmup at low volume, and a blunt line that mail from
  a cold domain lands in spam no matter how good the copy is.