Family logistics is a hidden workload: school letters in PDF and on paper, club chats, doctor appointments, pickups, chores, shopping. Most of it is unstructured text that someone must manually turn into calendar entries and lists. An LLM/VLM assistant reads these inputs, proposes events and tasks, spots clashes and sends a weekly briefing, while a human confirms anything that changes the shared calendar. Risk is low; the main concerns are children's data and who "owns" the household's mental load.
Problem & value
- Today: one or two adults act as the household "project manager": copying dates from letters and chat groups into calendars, remembering permission slips, balancing two or more work schedules. Tools are fragmented (calendar, notes, messenger, paper).
- AI value: extraction of dates, deadlines and required actions from messy sources; de-duplication across sources; clash and travel-time detection; fair, transparent task distribution; natural-language and voice capture ("dentist for Mia next Tuesday after school").
- Evidence: no robust, independent quantification of time saved exists that we could verify; vendors' claims should be treated as marketing and tested in your own household for a few weeks.
Variants
- Two-parent / patchwork families: shared calendars across households, handover days, custody schedules.
- Families with school-age kids: school portals (e.g. Schulportal, Elternportal, WebUntis, Klassenarbeiten dates), class-chat noise filtering.
- Multi-generational / care: coordinating care for elderly relatives, medication reminders (not medical advice; see PERS-014).
- Shared flats / WG: chore rotation, shared costs, cleaning plans.
- Wall-display hubs: a kitchen screen showing the day, fed by the assistant.
How it works
- sources (email, PDF, photo, chat export, voice)
- ingestion (OCR/VLM for paper and screenshots, ASR for voice notes)
- extraction (LLM: event, date/time, location, deadline, who is affected, required action)
- matching (de-duplicate vs. existing calendar, detect clashes and missing info)
- proposal card ("Add: Parent evening, 12 Nov 19:00, Room 204 — confirm?")
- human confirmation (one tap)
- write to shared calendar / task list
- reminders + weekly family briefing
- Key AI parts: structured extraction with a JSON schema (title, start, end, timezone, location, confidence); tool calling against calendar/task APIs; relative-date resolution ("next Friday") using the user's current date and time zone.
- Human-in-the-loop: all writes to a shared calendar are proposals; low-confidence dates and anything involving money, consent forms or medical items are never auto-actioned.
- Memory: a small household profile (members, schools, usual pickup times, allergies, preferences).
Data preparation
The data is the family's own: e-mails from school, daycare and clubs, PDFs, photos of notice-board letters, chat messages and voice notes, plus the existing calendars (Google, iCloud, Microsoft, CalDAV). Nothing sits in an ERP or DMS; the material is scattered across mailboxes, messengers, a paper pile and several calendars.
Typical gaps: phone photos of letters are skewed or blurry, many sources are unstructured chat, family members maintain separate calendars, and nobody has defined who owns which task or which calendar is the shared one.
Preparation steps: - Decide on one shared family calendar and one list app; migrate or subscribe the others. - Create a dedicated mailbox or forwarding address for school and club mail so the assistant sees only relevant messages. - Photograph letters flat in good light or use a scanner app with OCR (see PREP-006 for the principle, scaled down to a household). - Write a short household profile: family members, recurring routines, pickup rules, school holidays. - Limit access: give the assistant read access to selected mailboxes and write access only to a staging calendar, not the main one.
Effort share is low, roughly 10 to 20 percent of the total: an afternoon to set up, then the habit of forwarding or photographing letters. For tidy-up of legacy files see DOC-014.
Implementation paths
Which path for whom
- Private: Path A. Calendar and assistant features already built into Google, Apple or Microsoft ecosystems cover the task at near-zero cost and effort.
- Developer: Path C. A self-hosted stack (n8n, Ollama, CalDAV) keeps school letters, children's data and family chats at home and is a good fit for tinkerers.
Path A — Buy (recommended for most households)
- Built-in assistants: Google Gemini with Google Calendar/Tasks/Keep, Apple Intelligence and Siri with Reminders/Calendar, Microsoft Copilot with Outlook/To Do. Cheap and tightly integrated; data stays with the big platform.
- Family organiser apps: Cozi, Skylight (calendar display), Hearth Display, Famly/KiTa-Apps for daycare communication, FamilyWall, TimeTree; several add AI import of emails or photos. Check each product's current AI features and EU data location.
- EU/open options: Nextcloud (Calendar, Tasks, Deck; Germany/EU-based), Proton Calendar (Switzerland), Tuta Calendar (Germany), Mailbox.org.
- Choose A when: you want results in an evening and accept a US platform provider, or the school already uses a given app.
Path B — Build on cloud APIs
- Models: Claude, Gemini, GPT or Mistral (EU provider, La Plateforme) with structured outputs and function calling; vision input for photographed letters.
- Glue: Google Calendar API, Microsoft Graph, CalDAV; automation with n8n, Zapier, Make or Home Assistant automations; a Telegram/WhatsApp/Signal-style bot as the family interface.
- Pattern: forward school mails to a dedicated address → workflow extracts events → bot posts a confirm card in the family chat.
- Choose B when: you are a tinkerer and want custom rules (school-specific portals, custody calendars).
Path C — Self-host (open source)
- Calendar/tasks: Nextcloud, Radicale (CalDAV), Vikunja, Home Assistant calendar and to-do integrations; Mealie or Tandoor for recipes and shopping lists.
- Models: small open-weight LLMs (e.g. Qwen, Llama, Mistral, Gemma families) served via Ollama or llama.cpp; local Whisper-class ASR for voice notes; a small VLM or OCR (Tesseract, PaddleOCR) for paper letters.
- Hardware: HW1 suffices for ≤8B models and extraction of short texts (mini-PC, Apple Silicon, Raspberry Pi 5 for orchestration only); HW2 for better photo/PDF understanding. See PERS-005.
- Effort: a weekend to set up, then occasional upkeep.
Path D — Custom training
Not worthwhile for a household. Prompting plus a few example letters is enough.
State of the art (2026)
- General assistants now ship agentic features that act on calendars, mail and tasks; reliable extraction of dates from clean text is largely solved.
- Failure modes: wrong relative dates and time zones, mixing up children with similar names, misreading handwritten or low-quality photos, silently dropping "bring X" instructions, hallucinated locations, and duplicate events when several parents forward the same mail.
- Fairness of chore allocation is a social, not technical, problem; AI can make the load visible (who did what) but should not impose decisions.
- Prompt injection: an email or PDF can contain hidden instructions; an assistant with calendar/mail write access should show a diff before acting and never send outgoing messages without confirmation.
- No standard benchmark for "family logistics" exists; evaluate on your own 20–30 real letters.
Shared resources & combinations
- Same stack: email/doc extraction as in AGENT-001 and DOC-003; letter understanding PERS-008; voice capture shares ASR with PERS-017.
- Natural combinations: PERS-004 (meal plan → shopping list), PERS-003 (family trips), PERS-011 (shopping), PERS-019 (child safety and screen time), PERS-005 (host everything locally).
- Workflow engine (n8n/Home Assistant) is shared with AGENT-002.
Effort, cost & KPIs
- Path A: hours; subscription cost often free to low monthly fee.
- Path B: a weekend to a few weeks; API cost for a family's volume is typically small (a few euros per month at most, depending on model and images).
- Path C: a weekend plus upkeep; hardware you may already own.
- KPIs (self-assessed): missed deadlines/appointments per month, share of proposals accepted without edit, time spent on weekly planning, perceived fairness of chores (ask all members).
People & change
Here the "organisation" is the household. Involve all family members (end-users) before starting: agree which sources the assistant may read, who confirms calendar entries and who gets reminders. Children's data (names, schools, health appointments) is sensitive, so talk with older children about what is processed and keep younger children's details minimal. Schools, clubs and daycare (customers/affected parties) need no involvement, but do not forward their group chats that contain other parents' personal data into cloud tools without thinking about it.
Works council: not applicable, as this is private use. If a parent uses a work account or work calendar, keep employer data out of the household assistant and respect the employer's AI policy (GUIDE-007). A DPO or DPIA is not required for purely personal use (household exemption, Art. 2(2)(c) GDPR), but the tool provider remains a processor of your data, so check where it is hosted.
AI literacy: the AI Act asks providers and deployers to support literacy; privately, a short family conversation about errors (wrong dates, misread handwriting) is enough.
Typical concern: "who carries the mental load?" Address it by distributing tasks visibly and not letting the assistant silently decide. Keep humans in control: proposals first, confirmation by a person, weekly review of the briefing. See GUIDE-008 for acceptance tips and GUIDE-005 for data protection basics.
Risks, regulation & compliance
- EU AI Act: minimal risk for private household use. Do not use emotion recognition or covert monitoring features; Art. 5 prohibitions (e.g. emotion inference in workplace and education settings, manipulative or exploitative techniques) apply to providers and deployers of AI systems; private household use is largely outside deployer duties, but products offering "monitoring" of family members deserve scrutiny. AI literacy (Art. 4) applies to professional deployers, not private persons.
- GDPR: processing in a purely personal or household activity is outside GDPR (Art. 2(2)(c), Recital 18), but providers of the tools are controllers/processors and must comply. Schools, clubs and employers that deploy such tools are not exempt. Children's data deserves extra care (Recital 38); prefer EU hosting, no training on your data, and data deletion options.
- Third-party data: forwarded chats contain other parents' and children's names; share minimally and strip what is not needed.
- Security: least-privilege OAuth scopes (calendar only), no auto-send, separate mailbox for forwarding, review of connected apps.
- Social risk: surveillance of partners or teenagers via location or message features can harm trust and may be unlawful; agree rules openly.
Getting started
- List your sources (school mail, club chat, paper) and pick the single most painful one.
- Create a dedicated shared calendar and a forwarding address; keep personal calendars separate.
- Test an assistant (Path A) on 20 past letters; compare extracted dates with the truth.
- Require confirmation for every calendar write for the first month, then relax for low-risk items.
- Hold a short monthly family review: what worked, who feels overloaded, what data to delete.
Sources
- Regulation (EU) 2016/679 (GDPR), Art. 2 and Recitals 18, 38 — https://eur-lex.europa.eu/eli/reg/2016/679/oj
- Regulation (EU) 2024/1689 (AI Act) — https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- Google Calendar API documentation — https://developers.google.com/calendar/api
- Nextcloud Calendar and Tasks — https://nextcloud.com/calendar/
- Home Assistant calendar integration — https://www.home-assistant.io/integrations/calendar/
- Radicale CalDAV server — https://radicale.org/
- Ollama — https://ollama.com/
- OWASP Top 10 for LLM Applications (prompt injection) — https://genai.owasp.org/llm-top-10/
- Cozi Family Organizer — https://www.cozi.com/