OpenKB

OpenKB is an open-source command-line system from Vectify AI, the company behind PageIndex. It uses LLMs to compile raw documents into a structured, interlinked wiki-style knowledge base, with PageIndex's vectorless, reasoning-based retrieval handling long documents. It is released under the Apache 2.0 license, and the latest listed release is v0.4.5.

What It Is

OpenKB turns a folder of source material into a persistent wiki of Markdown files. Users initialize a knowledge base, add files, directories or URLs, and the LLM generates summaries, concept pages, entity pages (people, orgs, places, products) and cross-links. The README describes the approach as based on a concept described by Andrej Karpathy, and contrasts it with traditional RAG: knowledge is compiled once and kept current rather than rediscovered on every query.

How the Workflow Runs

Short documents are converted to Markdown with markitdown and read in full by the LLM. PDFs of 20 pages or more are processed by PageIndex into a hierarchical tree index with summaries, and the LLM reads the tree instead of the full text. Adding a document generates a summary, updates existing concept and entity pages, and refreshes the index and log. Supported inputs include PDF, Word, Markdown, PowerPoint, HTML, Excel, CSV, text and URLs.

Outputs and Generators

On top of the wiki, generators produce cited answers (openkb query), multi-turn chat sessions, an interactive knowledge graph, single-file HTML slide decks, and portable agent skills via the Skill Factory for Claude Code, Codex and Gemini CLI. Wiki pages follow the Google Open Knowledge Format and open in Obsidian. A bundled Knowledge Workbench web UI and a FastAPI REST API are available through the web extra.

Setup and Model Access

Install with pip, then configure any LiteLLM-supported provider such as OpenAI, Claude or Gemini. PageIndex runs locally by default; PageIndex Cloud can optionally be enabled with an API key for OCR on scanned PDFs, faster structure generation and scalable indexing.