OpenKB Launches Wiki-Style Knowledge Base Without Chunking or Vectors
VectifyAI has released OpenKB — an open-source knowledge base system inspired by Andrej Karpathy's idea to treat knowledge bases more like wikis than vector databases. Rather than chunking and embedding, OpenKB compiles raw documents into connected wiki structures using LLMs, with knowledge accumulating across queries. Features include automatic summaries, cross-document links, contradiction and gap detection, change-watching, and Obsidian-compatible Markdown. Long PDFs use PageIndex, a tree-style document index without vectors. One new file can update up to 15 wiki pages.
Why It Matters
OpenKB represents a structural alternative to the chunking-and-embedding RAG paradigm — particularly relevant as the 2026 practitioner conversation shifts from "how to build RAG" to "where it breaks under real workloads."