Kwipu
Kwipu turns a folder of Markdown notes, including an Obsidian vault, into a local property-graph retrieval-augmented generation index that can be queried from a terminal, a 3D web interface, or an MCP client.
Kwipu builds a property graph from Markdown, text, PDF and Word documents, extracting semantic relations with an LLM and structural relations from wikilinks and YAML front matter. It answers questions by combining vector similarity, BM25 keyword search, temporal metadata and optional synonym retrieval, grounding every answer in the source chunks it cites, and it watches the source folder to keep its index current.
Results can be explored through an interactive 3D graph in a web interface, asked directly from a terminal that keeps watching for note changes, or reached from an AI client over MCP. It connects to an Ollama-compatible endpoint and supports English, Italian, French, German, Spanish and Portuguese source material.
Kwipu can run fully locally with a local Ollama model, but its intentional default model is a cloud-hosted one that sends document chunks, retrieved context and questions to its provider, so a user must choose cloud or local before indexing. It is written in Python.