Practical overview
When it fits
Use Kwipu when you want a question-answering layer over an existing Obsidian vault or document folder, especially when tracing citations, relationships or changes across source material matters.
Who it is for
- Obsidian users who want to ask questions across a vault and follow relationships between notes.
- People organizing project or personal knowledge in ordinary document folders.
- Developers exploring local graph retrieval or connecting a knowledge index to an MCP client.
Documented uses
- Ask questions across notes and documents and inspect cited source material.
- Find relationships expressed through wikilinks, YAML frontmatter and semantic extraction.
- Combine vector, BM25 and temporal retrieval to locate relevant information.
- Explore entities, documents and relations in a 3D graph.
- Keep persisted index storage updated as a watched source folder changes.
Requirements & limits
Kwipu is presented as a self-hosted application with a public repository under the MIT license. The documented setup requires Python 3.12 or later and Ollama; Node.js and npm are needed for the web interface.
- Kwipu also lists .txt, .pdf and .docx as supported input formats, beyond Markdown.
- Retrieval combines vector similarity, BM25 and temporal metadata; synonym retrieval is optional.
- Answers are grounded in source chunks and return citations.
- The default cloud LLM may transmit document chunks, retrieved context and questions to its provider; fully local operation depends on selecting local LLM and embedding models.
- Changing the embedding model requires building a new storage index.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up Kwipu for terminal questions
Install Kwipu, add a document folder and start the documented terminal workflow.
Before you start
- Git or a downloaded repository ZIP
- Python 3.12 or later
- Ollama
- A folder containing .md, .txt, .pdf or .docx documents
Steps
Clone the Kwipu repository and change into its directory.
Create and activate a Python virtual environment, then install the packages from bridge/requirements.txt.
Place documents in the configured knowledge directory and make the Ollama LLM and embedding models available; choose local or cloud model settings before indexing.
Set the documented environment variables in the terminal, then run python .\geode_graph.py --fast and wait for the graph-built or graph-loaded message.
Enter a question at the terminal prompt and review the grounded answer; enter exit to finish.
Expected outcome
A terminal session for asking questions over the indexed documents, with answers grounded in source material.
Guide limitations
- The documented commands use Windows PowerShell; the README points to Linux and macOS equivalents.
- The default LLM can send document chunks, retrieved context and questions to its provider; local inference requires local LLM and embedding models.
- Keep the indexer process running if you want it to watch the knowledge folder for changes.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 9 October 2026
- Original website title
- benmaster82/Kwipu · Official homepage title · retained in its source language