Immich is a leading open-source option for: Google Photos–style backup, albums, and search on hardware you control. The immich-app/immich monorepo combines a server API, web UI, Flutter mobile apps, and a machine-learning stack for faces, objects, and CLIP-style search—aimed at people who will trade vendor convenience for ops responsibility.
Below is a concise overview, not a substitute for docs.immich.app.
TL;DR
| Question | Short answer |
|---|---|
| What? | Self-hosted photo/video library: mobile backup, web gallery, albums, map, faces, smart search |
| License | AGPL-3.0—read before commercial hosting or forking for networked services |
| Stack | NestJS server (per repo), SvelteKit web, Flutter mobile, ML container for models |
| Deploy | Docker-style stacks documented upstream (Postgres, Redis, ML service—see docs) |
| Your job | Backups (project README stresses 3-2-1), TLS, upgrades, storage planning |
| Links | immich.app · docs.immich.app · api.immich.app |
Features that matter in production
From the public README / feature matrix:
- Mobile: auto-backup when opened, background backup, dedup, RAW, Live/Motion Photo, offline-friendly gallery patterns.
- Library: albums, shared albums, partner sharing, public links, archive, favorites, stacked photos, folder view.
- Discovery: timeline, global map, “memories” (on-this-day style).
- Search: metadata; object/scene; facial recognition and clustering; CLIP semantic search where ML indexing is enabled.
- Admin: user management and API keys are web-first in their table; mobile vs web parity for tags and some views varies—check the current matrix before promising stakeholders.
Why the architecture matters for capacity
Immich splits responsibilities like a modern cloud product, but you run every tier:
- Core API — auth, libraries, jobs, storage layout.
- Clients — web and mobile talk to the same API.
- Machine learning — heavy inference (faces, embeddings, detection) off the interactive path.
Planning is not only disk: large libraries need CPU (or GPU, depending on how you run ML) for thumbnails, transcodes, and background ML queues—profile on your hardware.
AGPL-3 in practice
AGPL-3 can impose source-sharing obligations when you modify the app and let users interact over the network. Most home labs running upstream images unmodified have a different risk profile than hosting providers or white-label products. Get legal review before betting a business on a fork or a re-skinned multi-tenant Immich.
Related on ExplainX
- Sim (Sim Studio): open-source agent workflows and self-hosted ops
- Hermes Agent: self-hosted Nous Research CLI + Telegram
- What is MCP? Model Context Protocol guide
Sources
- Site: immich.app
- Docs: docs.immich.app
- Repository: github.com/immich-app/immich
- Roadmap: immich.app/roadmap
GitHub stars, release tags, and feature matrices change often. Treat this as May 5, 2026 context and verify versions before upgrades.