# AGENTS.md ## Project Overview MediaLore is a **self-hosted media library browser** for images and videos. It provides a web UI for browsing, searching, tagging, and viewing media with thumbnail previews and doom-scroll/TikTok-style consumption. **Monorepo** with two independent apps: - `backend/` — Python FastAPI server - `frontend/` — React SPA (TypeScript, Vite) ## Tech Stack | Layer | Technology | |-------------|---------------------------------------------------------| | Backend | Python 3.12+, FastAPI, Uvicorn | | Database | SQLite via SQLAlchemy (async aiosqlite), WAL mode | | Migrations | Alembic | | Thumbnails | Pillow (images), ffmpeg (videos) — generated on-demand | | File Watch | watchdog — live filesystem monitoring | | Frontend | React 19, TypeScript 6, Vite 8, React Router 7, TanStack Query v5 | | Infra | Docker Compose (backend: Python 3.12-slim, frontend: Nginx Alpine) | ## Directory Structure ``` medialore-web-app/ ├── docker-compose.yml # Orchestration (backend, frontend, volumes) ├── .env.example # MEDIA_ROOT env var template ├── backend/ │ ├── Dockerfile │ ├── pyproject.toml # Python package config (hatchling build) │ ├── alembic.ini │ ├── alembic/ # DB migrations │ │ ├── env.py │ │ └── versions/ │ └── app/ │ ├── main.py # FastAPI app, CORS, lifespan, router registration │ ├── config.py # Pydantic Settings (DATABASE_URL, MEDIA_ROOT, THUMBNAIL_DIR) │ ├── database.py # SQLAlchemy async engine, session, WAL pragma │ ├── models.py # ORM models: Library, MediaItem, Tag, media_item_tags │ ├── schemas.py # Pydantic request/response schemas │ ├── routers/ │ │ ├── libraries.py # Library CRUD, browse, doom-scroll, scan-status, rescan │ │ ├── media.py # Media item get, file serve, thumbnail, tag assignment │ │ ├── tags.py # Tag CRUD, grouped by category │ │ └── search.py # Fuzzy filename + tag search │ └── services/ │ ├── scanner.py # Dir walk, SHA-256 hashing, moved file detection │ ├── watcher.py # watchdog event handler (create/move/delete) │ └── thumbnails.py # Pillow + ffmpeg thumbnail generation └── frontend/ ├── Dockerfile # Multi-stage: Node build + Nginx serve ├── nginx.conf # SPA fallback, /api proxy to backend ├── package.json ├── tsconfig.json # Root references ├── tsconfig.app.json # App TS config ├── tsconfig.node.json # Vite config TS config ├── vite.config.ts # Vite + React plugin, dev proxy /api -> localhost:8000 ├── eslint.config.js # Flat config: JS + TS + React hooks + refresh ├── index.html ├── public/ │ ├── favicon.svg │ └── icons.svg └── src/ ├── main.tsx # React entry point ├── App.tsx # Router, QueryClient, sidebar, theme, responsive layout ├── index.css # CSS custom properties for light/dark themes ├── api/ │ └── client.ts # Typed fetch wrapper; all endpoint functions ├── pages/ │ ├── SearchPage.tsx │ ├── BrowserPage.tsx │ ├── SettingsPage.tsx ├── components/ │ ├── FileBrowser/FileBrowser.tsx │ ├── MediaViewer/MediaViewer.tsx │ ├── DoomScrollViewer/DoomScrollViewer.tsx │ └── TagPanel/TagPanel.tsx └── assets/ ├── hero.png └── vite.svg ``` ## Common Commands ### Backend ```bash # Install (from backend/) pip install -e . # Run dev server cd backend && uvicorn app.main:app --reload --port 8000 # Generate a new Alembic migration cd backend && alembic revision --autogenerate -m "description" # Apply migrations cd backend && alembic upgrade head ``` ### Frontend ```bash # Install dependencies (from frontend/) npm install # Dev server (proxies /api to localhost:8000) npm run dev # Production build npm run build # Lint npm run lint # Preview production build npm run preview ``` ### Docker ```bash docker compose up -d ``` ## Environment Variables | Variable | Default | Purpose | |------------------|-------------------------------------|------------------------------| | `DATABASE_URL` | `sqlite+aiosqlite:////data/medialore.db` | SQLite connection string | | `MEDIA_ROOT` | `/media` | Root path for media libraries | | `THUMBNAIL_DIR` | `/data/thumbnails` | Cached thumbnail storage | Pydantic Settings reads from `.env` at startup (via `model_config = {"env_file": ".env"}`). For Docker, these are set in `docker-compose.yml`. ## Database SQLite with **WAL mode** and a 10-second busy timeout (see `app/database.py:14-18`). Tables are auto-created at startup via `Base.metadata.create_all` in the FastAPI lifespan. ### Schema - **libraries**: `id`, `name`, `path` (unique) - **media_items**: `id`, `library_id` (FK), `rel_path`, `filename`, `file_hash` (SHA-256), `media_type` ("image" or "video"), `size_bytes`, `missing`, `created_at`, `updated_at` — unique on `(library_id, rel_path)` - **tags**: `id`, `name`, `category` — unique on `(name, category)` - **media_item_tags**: `media_item_id` (FK), `tag_id` (FK), composite PK **Migrations**: Use Alembic. The initial migration is at `backend/alembic/versions/0001_initial_schema.py`. ## API Conventions All endpoints are prefixed with `/api`. All responses are JSON except `/api/media/:id/file` (binary stream) and `/api/media/:id/thumbnail` (JPEG). ### Async Database Access Routers use FastAPI's `Depends` injection with the `get_db` async generator from `database.py`: ```python from app.database import get_db @router.get("/...") async def endpoint(db: AsyncSession = Depends(get_db)): ... ``` ### Frontend API Client All API calls go through the typed `api` object in `frontend/src/api/client.ts`. Use `useQuery` / `useMutation` from TanStack Query. Never call `fetch` directly from components. ```ts import { api } from "../api/client"; import { useQuery } from "@tanstack/react-query"; const { data } = useQuery({ queryKey: ["libraries", id, "browse", path], queryFn: () => api.libraries.browse(id, path), }); ``` ## Code Patterns & Conventions ### Backend (Python) - **No linter configured** — follow the existing style (4-space indent, snake_case) - Models use SQLAlchemy 2.0-style `mapped_column` (not `Column()`) - Schemas use Pydantic `BaseModel` - Service modules handle business logic; routers handle HTTP concerns - Background scanning runs via `asyncio.create_task`; watchers run in separate threads via `watchdog.Observer` - Logging: use `logging.getLogger(__name__)` — the `app` namespace has a StreamHandler configured in `main.py` ### Frontend (TypeScript/React) - **ESLint flat config** with strict TypeScript-ESLint rules - Inline `style={{ ... }}` objects for CSS (no CSS modules, no Tailwind) - TypeScript interfaces for all API types live in `client.ts` - Components use `useState`/`useEffect`/`useQuery` hooks, not class components - Theme state persisted in `localStorage` key `"theme"`, applied as `data-theme` attribute - Responsive via `window.matchMedia("(max-width: 767px)")` listener - File/thumbnail URLs constructed with `api.media.fileUrl(id)` and `api.media.thumbnailUrl(id)` (not fetched) ### TypeScript Configuration - `tsconfig.app.json`: strict mode, `noUnusedLocals`, `noUnusedParameters`, `erasableSyntaxOnly` - `tsconfig.node.json`: for `vite.config.ts` only - Build command: `tsc -b && vite build` (type-check via project references) ## Testing **There are no test files in this project.** No test framework is configured. When adding tests: - Frontend: consider Vitest (pairs with Vite) + React Testing Library - Backend: consider pytest + pytest-asyncio + httpx (for FastAPI TestClient) - No existing test commands in CI, so add them and consider a GitHub Action or similar ## Docker / Deployment - `docker compose up -d` starts both services - Backend runs on port 8000 (internal), frontend Nginx on port 80 (mapped to 8085 on host) - Nginx proxies `/api/` to `http://backend:8000` - Media paths from the host are volume-mounted into the container at `/media` - Persistent data (DB, thumbnails) stored in Docker volume `medialore-data` ## Important Notes - **No auth/authentication** — CORS allows all origins, no login system - Thumbnails are generated on first request and cached to disk (`THUMBNAIL_DIR`) - The scanner computes SHA-256 hashes to detect moved/renamed files and avoid re-processing - File watcher uses watchdog with per-library `PollingObserver` (necessary for Docker volumes) - `media_items.missing` flag is set when a scanned file no longer exists on disk - The frontend `build` command runs `tsc -b` first, which checks types across project references — type errors will block the build