1 Vision
Forgejo Actions edited this page 2026-09-25 16:38:07 +00:00

Memories Product Vision

Working Intent

Memories is a privacy-first, self-hosted/local-first photo and video library for a household. It should make personal media durable, browsable, and understandable without requiring trust in a cloud provider.

Product Goals

  • Preserve original photos and videos without modifying their bytes.
  • Make a family media archive easy to browse by time, album/event, and metadata.
  • Support family memory collaboration through shared libraries, group-specific sharing policies, and a durable people/face model with provenance.
  • Keep core functionality local/self-hosted and inspectable.
  • Make backup, restore, and migration understandable from day one.
  • Provide a polished, responsive web experience for the core archive flows using a coherent design system. Immich is the primary UI/UX quality and interaction reference, while Memories keeps its own architecture, product semantics, branding, and assets.
  • Design the initial architecture toward the intended long-term domain model while delivering it in maintainable, testable increments rather than chasing Immich feature parity.

Non-Goals for Early Versions

  • Full Immich compatibility or complete feature parity.
  • Dependence on Immich's implementation details, backend architecture, product semantics, or assets.
  • Cloud-hosted multi-tenant SaaS.
  • Generic advanced ML features as a v1 requirement. The local people/face pipeline is treated separately as core family-memory infrastructure when included by spec.
  • Public social-network style sharing.
  • Complex distributed storage before local storage boundaries are clear. Distributed workers for long-running local processing may be part of the architecture when they do not own authoritative media or metadata state.

Initial Product Questions

  1. What is the minimum acceptable mobile ingestion story for v1?
  2. Which metadata must be editable by the app, and which should remain read-only/indexed?
  3. What export/escape hatch is required before Christian trusts the system with real photos?
  4. Which parts of the family people/face workflow must be user-facing in the first release, and which only need stable model/API support initially?
  5. Which sharing-group defaults should be provided for a high-trust family group versus other groups?
  6. Which family/people collaboration workflows need the same UI polish as the core archive flows in the first release?

Architecture Questions to Decide Intentionally

  • Original file ownership and directory layout.
  • Metadata source of truth: database only, sidecar files, embedded metadata, or hybrid.
  • Thumbnail/cache invalidation and regeneration model.
  • Background job model for indexing, hashing, metadata extraction, thumbnails, and video previews.
  • API boundary between web UI, ingestion clients, and background workers.
  • Authorization boundaries for media owners, sharing groups, metadata sharing policies, and group-created face/person annotations.
  • People/face model boundaries: owner person records, group person projections, merge policy, approval workflow, and provenance.
  • Backup/restore guarantees and operational runbook.
  • Frontend design system and API contract boundaries for a polished but maintainable web UI.