Karakeep is a self-hostable bookmarking and read-it-later application. It supports saving links, notes, images, PDFs, and videos, with AI-powered auto-tagging, summarization, and semantic search. The system consists of a Next.js web app, an Expo React Native mobile app, a browser extension (Chrome/Firefox/Safari), a CLI, a Model Context Protocol (MCP) server, and a set of background workers (crawler, embeddings, feeds, webhooks, rule engine, backups). The backend uses tRPC for API, Drizzle ORM with SQLite, and a plugin system for swapping infrastructure (asset stores, queues, search, rate limiting). The architecture is monorepo with shared packages for types, server utilities, and React hooks. It is actively maintained with frequent releases and a large community.
BooruDatasetTagManager is a Windows Forms desktop application (C# .NET) for editing tags in image datasets used to train Stable Diffusion hypernetworks, embeddings, and LoRAs. It loads a folder of images with associated .txt tag files, provides a rich UI for tag editing (add, delete, reorder, weight, undo/redo), batch operations, filtering, translation, and auto-tagging via a companion Python AI server (Flask/FastAPI). The Python server supports multiple vision-language models (BLIP, Florence2, JoyCaption, Qwen, etc.) for captioning and tagging, plus background removal and translation. The architecture is a monolithic WinForms client with a separate AI inference server, communicating over HTTP. The codebase is functional and feature-rich but shows signs of monolithic design (large forms, static global state) and some deprecated practices (BinaryFormatter). Several components, such as the universal diffusion metadata parser (Metadata.cs) and the WebP codec wrapper, are well-isolated and potentially reusable in other projects.