A modular, AI + file home server designed to run fully on your local machine. LORE combines storage, search, media intelligence, and local LLM capabilities into one coherent ecosystem.
LORE is built around a set of focused components, each named after figures or artifacts from Norse mythology. Every module has a specific job but works together to form a unified local AI stack.
Bifrost → API Gateway & Router
Odin → Local LLM Engine
Mimir → Storage System
├── Orion → Media & Photo Intelligence
└── Saraswati → Metadata Database
Freya → Frontend Android App
Psyche → Vector Embedding Database
Bifrost is the entry point to the entire system. It handles routing, authentication, and acts as the API gateway. All user requests pass through Bifrost before reaching other services.
Odin runs one or more local language models.
- Supports a lightweight daily‑chat model and a heavy reasoning model.
- Handles text, file, and image inputs.
- Integrates directly with Mimir for file access.
- Optionally integrates with Psyche for vector search.
Mimir manages all local files with permission‑scoped access.
- Upload, download, rename, delete
- File sharing via permission‑linked tokens
- Embeddings generated from filenames, notes, and content
- Houses two submodules:
Handles all media operations, especially photos.
- Uses YOLO/CLIP‑style models for tagging
- Can perform object detection and possible facial recognition
- Sends metadata to Saraswati
A SQLite‑based metadata store.
- Tracks permissions, timestamps, file relationships
- Stores metadata for both Mimir and Orion
Freya is the UI layer of LORE, built using Kotlin and Jetpack Compose.
- Mobile-first interface for browsing files, media, and chatting with Odin.
- Connects directly to Bifrost.
- Provides a polished, modern experience for interacting with the whole system.
Stores and retrieves embeddings for semantic search.
- Implemented with ChromaDB
- Used by Odin and Mimir for similarity search
- Backend: FastAPI
- DB: SQLite for metadata, ChromaDB for vector embeddings
- Models: Llama 4 (planned), YOLO/CLIP‑style models for images
- Storage: Local filesystem
- Fully local AI experience
- Modular components that can be swapped or extended
- Minimal external dependencies
- Smooth file + LLM integration
(Placeholder — fill once installation steps are finalized)
- Automatic model switching
- User roles & granular permissions
- Web UI for browsing Mimir & Orion
- Real‑time media processing
- Cross‑module event system
Pull requests and feature ideas are welcome once the public repo is live.
Designed and built with a blend of engineering chaos and mythological inspiration.