Omoide

Omoide is a self-hosted, offline-first photo and video library that organizes media locally using on-device AI models. It focuses on privacy and longevity by keeping all processing and search on the user’s machine, and can be run as a desktop app or in Docker for servers and NAS devices.

Key Features

• 100% local / offline-first media management (no cloud required).
• Face detection, embedding and clustering with review/merge tools for people management.
• Semantic (natural-language) image search powered by OpenCLIP embeddings.
• Auto-tagging, perceptual duplicate detection, and video scene extraction for faster curation.
• Map view with EXIF extraction and GPS editing; co-appearance graph to visualize relationships between people.
• Multiple profiles, read-only mode for sharing, and background task management for large libraries.

Use Cases

• Personal archival and private photo/video libraries where users want local AI-powered search and face grouping.
• Photographers or hobbyists organizing large collections on a NAS or local server (Docker support).
• Small galleries or archivists needing an offline, searchable presentation/preview of media with privacy controls.

Limitations and Considerations

• License: distributed under the PolyForm Noncommercial License 1.0.0, which restricts commercial use; review the license before deploying in commercial contexts.
• Resource & setup: initial model downloads and AI processing can require significant CPU/GPU resources; FFmpeg and Python toolchain are required for desktop builds and media processing. Performance and required hardware depend on chosen models (CPU-only runs will be slower).
• Community & support: actively maintained as a passion project; community discussion and feedback appear on self-hosting forums and social channels rather than a dedicated commercial support portal.

Omoide provides a privacy-focused, local-first alternative for people who want Google/Apple Photos–like discovery features without cloud dependencies. It is suited for users who can accommodate model downloads and local compute demands and who accept the project’s noncommercial license terms.