🪴 Brain

OpenKB

3 min read

OpenKB

Open LLM Knowledge Base — an open-source CLI (VectifyAI) that compiles raw documents into a structured, interlinked wiki-style knowledge base using LLMs. It is a direct, packaged implementation of the Karpathy LLM Wiki pattern: instead of re-deriving knowledge on every query (traditional RAG), it compiles knowledge once into a persistent wiki and keeps it current. Apache-2.0.

This is the same architecture this vault runs by hand — a relevant reference for Brain and Building an AI Second Brain.

Description

Two layers:

  • Wiki foundation — compiles and maintains the knowledge. On add, the LLM generates a summary page, reads existing concept/entity pages, synthesizes cross-document concepts, creates/updates entity pages (people, orgs, places, products — auto-extracted, kept in sync), and updates the index + log. One source can touch 10–15 wiki pages; knowledge accumulates rather than sitting isolated.
  • Generators — turn the wiki into output: query/chat (grounded, cited answers), visualize (self-contained 3D/mind-map/radial knowledge graph HTML), and the Skill Factory.

Retrieval is vectorless — powered by PageIndex, which builds a hierarchical tree index over long PDFs (≥20 pages) and lets the LLM reason over the tree instead of chunk-and-embed. No vector DB. Short docs are read in full via markitdown.

Wiki pages follow Google’s Open Knowledge Format spec and are plain .md with [[wikilinks]] — opens directly in Obsidian for graph view.

Download or use

Reasoning for

This vault is a hand-built version of exactly what OpenKB automates: a compiled, cross-linked wiki maintained by an LLM agent (CLAUDE.md schema, progressive-disclosure indexes, entity-style notes). OpenKB is the closest packaged tool — worth tracking for (a) its vectorless PageIndex retrieval as an alternative to embedding-based Graphify/LightRAG/CocoIndex, (b) the Skill Factory (openkb skill new) which distills redistributable Agent Skills from the wiki, mirroring the skills this repo ships, and (c) its OKF compliance.

🧩 Skill Factory

openkb skill new <name> "<intent>" distills an installable agent skill from any subset of the wiki — a SKILL.md + references/ + optional scripts/, plus an auto-updated marketplace.json for one-line install. Quality gates: structural validate, trigger-accuracy eval (does the description: actually fire?), and history/rollback versioning. Iterate inside openkb chat without re-running the pipeline.

Alternatives considered

  • Karpathy’s manual workflow — web clipper → .md, LLM compiles; OpenKB adds long-doc PageIndex, automatic entity extraction, broad format support, and CLI agent integration.
  • Graphify / LightRAG / CocoIndex — embedding- or graph-based RAG/indexing; OpenKB is the vectorless, persistent-wiki approach.
  • Open Notebook — NotebookLM-style chat-with-sources, not a compiled wiki.
  • OpenWiki — langchain-ai’s code-specific counterpart: compiles a wiki of your codebase for agents, kept fresh by a daily-PR GitHub Action.

Resources

Template: tool

Graph View

Backlinks