afrexai-email-to-calendar

name	afrexai-email-to-calendar
version	1.0.0
description	Extract calendar events, deadlines, action items, and follow-ups from emails. Works with any calendar provider (Google, Outlook, Apple, Notion, etc.). No external dependencies — pure agent intelligence. Use when the user forwards an email, asks to check inbox for events, or wants to extract structured scheduling data from any text.

Email → Calendar Extraction Engine

Turn emails into structured calendar events with zero missed deadlines.

Quick Start

When you receive an email (forwarded, pasted, or from inbox):

1. Parse — Extract every time-relevant item using the framework below
2. Classify — Score each item by type and confidence
3. Present — Show structured results with numbered selection
4. Create — Use the user's calendar tool to create confirmed events
5. Follow up — Track deadlines and send reminders

---

1. Extraction Framework

What to Look For

Scan every email for ALL of these categories:

Category	Signals	Priority
Hard Events	"meeting at", "call on", "event on", specific date+time	🔴 High
Deadlines	"due by", "submit before", "RSVP by", "register by", "expires"	🔴 High
Soft Events	"sometime next week", "let's meet soon", "planning for March"	🟡 Medium
Recurring	"every Monday", "weekly", "monthly", "standing meeting"	🟡 Medium
Action Items	"please review", "can you send", "follow up on", "action required"	🟡 Medium
Travel/Logistics	Flight numbers, hotel confirmations, check-in times, gate info	🔴 High
Implicit Deadlines	Event is Feb 20 → ticket deadline is likely 1-2 weeks before	🟡 Medium

Extraction Template

For each item found, extract:

- title: "Descriptive name (max 80 chars)"
  type: event | deadline | action_item | travel | recurring
  date: "YYYY-MM-DD"
  day_of_week: "Monday"  # Always include for verification
  time_start: "14:00"    # 24h format, default 09:00 if unclear
  time_end: "15:00"      # Default: start + 1h for meetings, all-day for deadlines
  timezone: "America/New_York"  # Extract from email headers or content
  is_all_day: false
  is_multi_day: false     # If true, include end_date
  end_date: null
  recurrence: null        # "weekly" | "biweekly" | "monthly" | "MWF" | custom RRULE
  location: null          # Physical address or video link
  url: null               # Registration link, event page, or action URL
  attendees: []           # Names/emails mentioned
  confidence: high | medium | low
   
   
      
         
       

Confidence Scoring

Confidence	Criteria
High	Explicit date + time + clear event type. E.g. "Meeting on Feb 15 at 2pm"
Medium	Date but no time, or time but approximate date. E.g. "next Tuesday afternoon"
Low	Vague reference. E.g. "we should catch up soon", "sometime in March"

Smart Defaults

• No time given for meeting → 09:00-10:00 (mark confidence: medium)
• No time given for deadline → 23:59 (end of day)
• No timezone → Use user's default timezone, note assumption
• "Morning" → 09:00, "Afternoon" → 14:00, "Evening" → 18:00, "EOD" → 17:00
• "Next week" → Following Monday (mark confidence: medium)
• Multi-day event → Set is_multi_day: true, include start and end dates

---

2. Email Classification

Before extracting, classify the email:

Email Type	How to Handle
Calendar notification (from [e-mail], outlook, etc.)	SKIP — these are responses to existing events
Newsletter/marketing	Extract only if contains relevant event dates
Personal/work email	Full extraction
Travel confirmation	Extract ALL logistics: flights, hotels, car rentals, check-ins
Meeting invite (ICS attachment or structured invite)	Extract directly, high confidence
Thread/reply	Only extract NEW events, not ones from quoted text
Forwarded email	Process the forwarded content, note original sender

Ignore Patterns (Skip These)

• Automated calendar responses (Accepted, Declined, Tentative)
• Unsubscribe confirmations
• Read receipts
• Auto-replies / Out of office
• Spam/promotional (unless user explicitly forwards it)

---

3. Presentation Format

Always present extracted items in this format:

📧 From: [sender] | Subject: [subject] | Date: [received date]

Found [N] calendar items:

1. 🔴 **Team Standup** — Mon Feb 17, 9:00-9:30 AM EST
   📍 Zoom (link in email) | 👥 Alice, Bob, Charlie
   🔁 Recurring: Every weekday
   ✅ Confidence: High

2. 🔴 **Project Deadline: Q1 Report** — Fri Feb 28, EOD
   ⚠️ ACTION REQUIRED: Submit report
   🔗 [Submission portal](url)
   ⏰ Suggested reminder: 3 days before
   ✅ Confidence: High

3. 🟡 **Team Lunch** — "sometime next week"
   📍 TBD
   ⚠️ Confidence: Medium — date needs confirmation

---
Reply with numbers to create (e.g. "1, 2"), "all", or "none".
Type "edit 3" to modify before creating.

Presentation Rules

1. Always show day of week — humans verify dates by day name
2. Group by date when >5 items
3. Flag conflicts — if new event overlaps existing calendar
4. Highlight deadlines with ⚠️ and days remaining
5. Show source quote for medium/low confidence items
6. Never auto-create without user confirmation

---

4. Calendar Creation

After user confirms, create events using their calendar tool:

Google Calendar (via gog or API)

gog calendar create \
  --title "Event Title" \
  --start "2026-02-17T09:00:00-05:00" \
  --end "2026-02-17T10:00:00-05:00" \
  --description "Extracted from email: [subject]" \
  --location "Zoom link or address"

Apple Calendar (via osascript)

osascript -e 'tell application "Calendar"
  tell calendar "Work"
    make new event with properties {summary:"Event Title", start date:date "Monday, February 17, 2026 at 9:00:00 AM", end date:date "Monday, February 17, 2026 at 10:00:00 AM", description:"Extracted from email", location:"Zoom"}
  end tell
end tell'

Notion / Other

• Format as structured data and use the appropriate API
• Or output as .ics file the user can import anywhere

ICS Export (Universal)

BEGIN:VCALENDAR
VERSION:2.0
BEGIN:VEVENT
DTSTART:20260217T090000
DTEND:20260217T100000
SUMMARY:Event Title
DESCRIPTION:Extracted from email
LOCATION:Zoom link
END:VEVENT
END:VCALENDAR

---

5. Duplicate Detection

Before creating any event, check for duplicates:

1. Search calendar for events on the same date with similar title (fuzzy match)
2. Check tracking file — maintain a log of created events:

// memory/email-calendar-log.json
{
  "created_events": [
    {
      "email_id": "msg-123",
      "email_subject": "Team Offsite",
      "event_title": "Team Offsite",
      "event_date": "2026-02-17",
      "calendar_event_id": "cal-456",
      "created_at": "2026-02-13T10:00:00Z"
    }
  ]
}

3. If duplicate found: Show user and ask — "This looks similar to [existing event]. Skip, update, or create anyway?"

---

6. Deadline & Reminder Engine

Deadline Patterns to Detect

Pattern	Example	Action
RSVP deadline	"RSVP by Feb 10"	Create reminder 3 days before
Registration	"Register by March 1"	Create reminder 1 week before
Early bird	"Early bird ends Feb 15"	Create reminder 2 days before
Ticket sales	"Tickets on sale until..."	Create reminder + calendar event
Submission	"Submit proposal by..."	Create reminder 3 days before
Expiration	"Offer expires..."	Create reminder 1 day before

Reminder Strategy

• >30 days away: Remind 1 week before
• 7-30 days away: Remind 3 days before
• <7 days away: Remind 1 day before
• Deadlines with URLs: Include the action URL in the reminder
• Create reminder as separate calendar event: "⚠️ DEADLINE: [action] for [event]"

---

7. Travel Email Handling

Travel confirmations get special treatment:

Extract ALL of these:

• ✈️ Flights: airline, flight #, departure/arrival times+airports, terminal, gate, confirmation #
• 🏨 Hotels: name, address, check-in/out times, confirmation #
• 🚗 Car rentals: company, pickup/dropoff times+locations, confirmation #
• 📋 Transfers: shuttle times, train bookings

Create these calendar events:

1. Flight departure — include terminal, gate, flight # in description
2. Flight arrival — for connecting flights too
3. Hotel check-in — with address and confirmation #
4. Hotel check-out — with reminder to pack
5. Car pickup/dropoff — with location details

Travel-specific reminders:

• Flight: 3 hours before (domestic), 4 hours before (international)
• Hotel check-out: Morning of departure
• Include all confirmation numbers in event descriptions

---

8. Batch Processing

When scanning an inbox for events:

1. Fetch unread emails (or emails from last N days)
2. Filter out noise — apply ignore patterns
3. Extract from each — run extraction framework
4. Deduplicate across emails — same event mentioned in multiple threads
5. Sort by date — nearest first
6. Present grouped summary:

📬 Inbox Scan: 47 unread → 12 with calendar items → 18 events found

THIS WEEK (Feb 13-19):
1. 🔴 Sprint Review — Thu Feb 13, 3:00 PM
2. 🔴 1:1 with Manager — Fri Feb 14, 10:00 AM
...

NEXT WEEK (Feb 20-26):
5. 🟡 Team Lunch — date TBD (mentioned in 2 emails)
...

DEADLINES:
⚠️ Q1 Report — Due Feb 28 (15 days) → [Submit here](url)
⚠️ Conference RSVP — Due Feb 20 (7 days) → [RSVP](url)

---

9. Edge Cases

Situation	How to Handle
Multiple timezones in one email	Extract each event in its stated timezone, convert to user's TZ for display
"TBD" or "TBA" times	Create all-day event, flag for follow-up
Cancelled events	Check if already in calendar → offer to delete
Rescheduled events	Find original → offer to update (not create new)
Recurring with exceptions	Note specific exception dates in description
Date ambiguity (02/03 = Feb 3 or Mar 2?)	Use email's locale/origin for MM/DD vs DD/MM, ask if unclear
Events in quoted/forwarded text	Only process if user explicitly forwarded it
Attachments with .ics files	Parse ICS directly — highest confidence source
"Save the date" emails	Create tentative event, mark as placeholder
Conference with multiple sessions	Extract all sessions as separate events with shared description

---

10. Session Memory

Track user preferences across sessions:

# memory/email-calendar-prefs.yaml
default_timezone: "America/New_York"
default_calendar: "Work"
default_reminder_minutes: 30
auto_create_patterns:
  - "standup"
  - "1:1"
ignore_patterns:
  - "newsletter"
  - "marketing"
preferred_format: "12h"  # or "24h"
travel_reminder_hours: 3

Update preferences when user corrects you or states a preference.