Building an LLM Wiki for Ditto: How a Mac Mini Became My Second Brain

Adam Fish, founder and CEO, describes building a Karpathy-style LLM wiki for Ditto on a Mac mini, using Claude Code for infrastructure and local hybrid search, to create a persistent knowledge base that replaces traditional RAG pipelines. This wiki organizes business knowledge into structured markdown pages, separating internal company operations from external marketing content, and employs a local hybrid search tool, qmd, for fast and private queries. Fish emphasizes the importance of context engineering over simple retrieval, arguing that Karpathy's LLM wiki pattern allows AI to maintain a deep, evolving understanding of a business by pre-structuring, pre-linking, and pre-summarizing knowledge. The setup involves two separate wikis for public and private data, each with different risk profiles, running continuously on a Mac mini with local file sync for access on other devices. Fish explains the decision to use Claude Code over OpenClaw for enhanced trust in handling sensitive company data, highlighting the potential of local-first AI for privacy and control. The template for building such a wiki is available on GitHub, allowing others to customize it for various domains like marketing, research, and customer intelligence.