Agentic Cloudflare Pipeline

Photo discovery — Apple Photos sqlite

The Photos library has a sqlite DB at a fixed path with face-recognition tables.

# Database path
DB=~/Pictures/Photos\ Library.photoslibrary/database/Photos.sqlite

# Find person clusters by name (multiple "Cynthia" clusters are common)
sqlite3 $DB "SELECT p.Z_PK, p.ZFULLNAME, COUNT(DISTINCT f.ZASSETFORFACE) as photos
  FROM ZPERSON p
  JOIN ZDETECTEDFACE f ON f.ZPERSONFORFACE = p.Z_PK
  JOIN ZASSET a ON a.Z_PK = f.ZASSETFORFACE
  WHERE p.ZFULLNAME LIKE '%Cynthia%' AND a.ZTRASHEDSTATE = 0
  GROUP BY p.Z_PK HAVING photos > 5 ORDER BY photos DESC"

Key tables and joins:

• ZPERSON — face clusters Apple has named or you've tagged. Z_PK = person id.
• ZDETECTEDFACE — every face Apple detected. ZPERSONFORFACE → person, ZASSETFORFACE → photo.
• ZASSET — the actual photo/video. ZKIND 0=photo, 1=video. ZTRASHEDSTATE 0=alive.

Apple's ML-ranked quality columns on ZASSET (huge for picking "the good ones" automatically):

• ZOVERALLAESTHETICSCORE · 0.0–1.0 · the master beauty score
• ZCURATIONSCORE · how often Apple includes it in Memories
• ZICONICSCORE · how prominent the face is (faces only)
• ZFAVORITE · 1 if you tapped heart
• ZLATITUDE / ZLONGITUDE · GPS, -180 = unknown
• ZDATECREATED · seconds since 2001-01-01 (add 978307200 to get unix epoch)

Asset materialization — bypass iCloud "Optimize Mac Storage"

Originals are often not on disk (iCloud-resident). But Apple keeps two on-disk variants in resources/derivatives/ that are perfect for web:

# Originals (often missing in iCloud-optimize mode)
~/Pictures/Photos\ Library.photoslibrary/originals/{0-F}/{UUID}.{ext}

# Web-friendly preview ~1024px (always local) — USE THIS
~/Pictures/Photos\ Library.photoslibrary/resources/derivatives/masters/{0-F}/{UUID}_*.jpeg

# Tiny thumbnail (fallback)
~/Pictures/Photos\ Library.photoslibrary/resources/derivatives/{0-F}/{UUID}_*.jpeg

Python copy with multi-tag aggregation:

import sqlite3, shutil, glob, os, json

# Get every photo containing at least one of 4 target person Z_PKs,
# aggregating ALL persons in each photo so the filter can do union logic
q = """
SELECT a.Z_PK, a.ZUUID, datetime(a.ZDATECREATED+978307200, 'unixepoch'),
       a.ZLATITUDE, a.ZLONGITUDE,
       a.ZOVERALLAESTHETICSCORE, a.ZFAVORITE,
       GROUP_CONCAT(DISTINCT f.ZPERSONFORFACE) as person_pks
FROM ZASSET a
JOIN ZDETECTEDFACE f ON f.ZASSETFORFACE = a.Z_PK
WHERE f.ZPERSONFORFACE IN (5903,26988,5858,5944)
  AND a.ZTRASHEDSTATE = 0 AND a.ZKIND = 0
GROUP BY a.Z_PK
"""
# ... then copy largest derivative for each UUID

Evernote .enex extraction (bonus visual gifts)

Evernote exports are XML with base64-embedded resources. Tonight, three .enex files contained the actual 2023 Mother's Day cards as PNGs:

import re, base64

with open(enex_path) as f: txt = f.read()
resources = re.findall(r'<resource>(.*?)</resource>', txt, re.S)
for r in resources:
    mime = re.search(r'<mime>([^<]+)', r).group(1)
    if 'image' not in mime: continue
    raw = base64.b64decode(re.search(r'<data[^>]*>(.*?)</data>', r, re.S).group(1).replace('\n',''))
    open(out_path, 'wb').write(raw)

Static site build — single index.html + JSON-driven render

Three files do everything:

• index.html — semantic structure, embedded CSS (Playfair Display via Google Fonts), inline JS at end
• assets/photos-manifest.json — array of { file, taken, moms[], lat, lng, score, favorite }
• assets/photos/all/*.jpeg — the copied derivatives

Filter logic — the cleanest multi-tag pattern I know:

// On each photo cell:
cell.dataset.moms = p.moms.join(" ");  // "cynthia jean"

// On chip click:
chip.addEventListener("click", () => {
  activeFilter = chip.dataset.mom;
  document.querySelectorAll(".cell").forEach(cell => {
    const moms = cell.dataset.moms.split(" ");
    cell.classList.toggle("hidden", activeFilter !== "all" && !moms.includes(activeFilter));
  });
});

Deploy to Cloudflare Pages

Authentication once via OAuth (one-time, browser popup):

npx wrangler login

Then create the project (only once per site) + deploy:

# Project create — gives you the *.pages.dev domain
(unset CF_API_TOKEN; unset CLOUDFLARE_API_TOKEN;
 npx wrangler pages project create mothers-day-2026 --production-branch=main)

# Deploy
(unset CF_API_TOKEN; unset CLOUDFLARE_API_TOKEN;
 npx wrangler pages deploy dist \
   --project-name=mothers-day-2026 \
   --commit-dirty=true \
   --branch=main)

Password gate via Pages Function

Cloudflare Pages auto-bundles any code under ./functions/ (sibling of the deploy dir, NOT inside it):

// functions/_middleware.js — runs on every request
export const onRequest = async (context) => {
  const { request, next } = context;
  const url = new URL(request.url);

  if (url.pathname === "/login") return handleLogin(request);

  const cookieHdr = request.headers.get("Cookie") || "";
  const authed = cookieHdr.split(";").some(c => c.trim() === "mday2026=you-are-welcome-here");

  if (authed) return next();
  return new Response(loginHtml(false), {
    headers: { "content-type": "text/html; charset=utf-8" }
  });
};