LLM Wiki setup: build a cited local knowledge base

This tutorial presents LLM Wiki as a persistent knowledge synthesis tool that turns source documents into an interlinked Markdown vault. The speaker contrasts this approach with repeated RAG retrieval and describes links back to locally stored sources.

Setup starts with an operating-system-specific binary from the official GitHub releases page. The walkthrough lists macOS, Linux and Windows 10 or 11, with Microsoft Edge WebView2 required on Windows. It recommends version 0.4.24 or later for a known Windows blank-screen problem. Processing can use OpenAI or Anthropic APIs, or Ollama for local inference. Separate models can handle ingestion and analytical queries. The speaker warns that API keys reside in plaintext in a local app state JSON file.

The research template creates an Obsidian-compatible vault. schema.md controls page structure, YAML front matter and wikilinks; purpose.md records research goals. AutoWatch queues PDFs and Markdown for ingestion, while SHA-256 hashes let the system skip unchanged sources.

The walkthrough then covers cited chat answers, retrieval that combines lexical, semantic and graph methods, and Louvain clustering in the graph view. It also describes a web clipper and a dedicated skill that lets terminal AI agents query the wiki through its built-in API.