Roadmap

What the Mediathek is meant to become, beyond what is built. An item here is a direction, not a promise. When work starts, it gets an ADR in adr/, and its diagrams appear in the gallery.

Archive: documents that belong to the family

Status: idea (2026-11-04). No code yet.

What

Besides photos and films, a family keeps papers that matter. The Archive should cover:

• the children's school work: scans, PDFs, drawings photographed;
• contracts, invoices, warranties and manuals;
• insurance cases with their evidence: photos of the damage, receipts, letters;
• certificates and records.

Each document is kept once, can be found again, and is kept as safely as the photos.

What already exists to build on

• The edge's scan already lists documents as their own kind: pdf, txt, doc, docx, csv, html and svg (engine.py, KINDS["document"]). Today they are listed but never opened. Fingerprinting covers only photos, RAW and films.
• Fingerprints, the bin and stores (ADR-0012) work for any file, so finding duplicates and cleaning a drive carry over unchanged.
• Vaults (ADR-0010) already decide who sees what.
• The knowledge graph (ADR-0008) already holds people, places and events, which documents can point to.
• Office documents and other files are published as their original (ADR-0014), and each can carry a summary in en/de/fr, a document type, its own date and the people and places it names. The edge's chat writes it as an AI draft, and a person checks it (ADR-0016).
• Photos and films take the same summaries, and the edge finds similar items for a selection by how they look, their text and what is known, so the chat can name and tag a group (ADR-0017).

Stages

1. Onboarding: fingerprint and remove duplicate documents too, as an option on the drive (opt-in, so a photo-only onboarding stays as fast as today).
2. Publish to a vault, with a preview (the first page) and the text: the PDF's text layer first, then OCR.
3. An Archive view in the Mediathek: by person (a child's school years), by subject (car, house, insurance), by date, with full-text search.
4. Cases: an insurance claim or a contract gathers documents and photos, with its dates and parties. In the graph it is a schema.org Event or CreativeWork, and its documents are DigitalDocuments.
5. Keep-until dates, for example invoices for 10 years and warranties until they end, with a reminder before something may go.

Open questions

• Sensitive papers (IDs, health, money): a separate, stricter vault? Encrypted at rest with a household key, so the cloud operator cannot read them?
• Children's work: who may see it, and from what age a child decides that for themselves.
• OCR: on the edge, so nothing is uploaded just to be read, or in the cloud later?
• Long-term formats: convert to PDF/A when published, or keep the original only?