Build a self-organizing personal knowledge base from your documents
Drop PDFs or web pages into the app, and an LLM reads the content to automatically create interconnected wiki pages. Instead of searching for answers from scratch with every query, this approach maintains and updates knowledge once it has been organized. You can watch your documents transform into a structured knowledge network without complex configuration.
Organize once and keep it maintained
Traditional Retrieval-Augmented Generation (RAG) methods require finding the original document and generating an answer for every question. This app works by having the LLM generate structured wiki pages as it reads documents, and then modifying or linking existing knowledge when new documents arrive. The resulting knowledge persists after restarting the app and is stored in a form that users can edit directly. It employs a two-step thinking process, analyzing content before generating pages, which results in documents with clear sources and logical structure.
Knowledge graph and automatic research features
Beyond simple document listing, it provides a knowledge graph that visualizes relationships between pages. It analyzes direct links, shared sources, and common neighbors to automatically cluster highly relevant pages and identify unexpected connections or knowledge gaps. It also includes a deep research feature that finds additional material via web search and automatically integrates it into the wiki for discovered gaps. Without requiring users to input search terms directly, the LLM understands the context to suggest optimal search topics and organizes the results.
Support for various document formats and browser extension
It supports various document formats including PDF, Word, Excel, PowerPoint, and e-books (EPUB/MOBI), as well as images and web clips. Installing the Chrome browser extension allows you to capture web pages with a single click and add them directly to the knowledge base. Images extracted from documents are described by a vision LLM and displayed alongside search results, with the ability to jump directly to their original location. All processing occurs locally, calling external LLM APIs only when necessary.
Installation and requirements
Pre-built binaries for macOS, Windows, and Linux can be downloaded and installed from the release page. To build from source, you need Node.js 20 or higher, Rust 1.88 or higher, and protoc. The browser extension is installed by selecting the extension/ folder via 'Load unpacked extension' in Chrome's developer mode. The app is built on Tauri v2 to support cross-platform environments, and automated builds are provided through GitHub Actions.
Pre-use considerations
This tool requires an LLM API key and supports various providers such as OpenAI, Anthropic, Google, and Ollama. Vector search is optional; if enabled, it requires LanceDB and an OpenAI-compatible endpoint. The README does not provide performance comparisons for specific LLM models or concrete cost data, so you should select a model that fits your usage pattern. Additionally, if your documents are sensitive, it is recommended to configure a local LLM (such as Ollama) to ensure data does not leave your system.
