AI Reader V2

AI Reader is an open source system that turns long Chinese novels into interactive knowledge graphs. It uses a large language model to extract characters, relationships, locations and events from books of 100 to 2,500 chapters, then generates four linked views automatically: a relationship graph, a procedurally generated world map, an event timeline and a character encyclopedia. It is aimed at readers, literature researchers and fans of long serialized fiction who want to follow sprawling casts and geographies. The relationship graph is force directed with category colored edges and path finding, built to stay legible for novels with more than 400 characters. The world map uses constraint solved layout with drawn terrain and animated character trajectories. The timeline tracks events chapter by chapter with character swimlanes, and the encyclopedia builds cross chapter profiles with alias resolution. Because models make systematic errors, extraction passes through a six layer validation pipeline covering place name rules, geographic rank correction, multi source voting, transitivity checks, hierarchy consolidation and a final model review of suspicious pairs. AI Reader is free. A live demo with analyzed editions of Journey to the West and Dream of the Red Chamber runs in the browser, and a desktop app for Apple Silicon Macs and Windows analyzes your own novels locally, keeping data on your machine. The desktop app requires Ollama to run models locally. The project is labeled an early preview and its author warns that results may contain errors and are not yet production ready. Its method is described in a paper accepted to the EMNLP 2026 demo track, with a companion benchmark published on Hugging Face.