Your documents deserve a wiki, not a vector search

What it does

OpenKB is a CLI tool that ingests documents—PDFs, directories, or URLs—and uses an LLM to build a structured wiki with summaries, concept pages, entities, and cross-references. Short documents are read in full; long PDFs are indexed by PageIndex into hierarchical trees so the LLM can reason over them without a vector database. Once compiled, the wiki/ directory serves as a persistent foundation for querying, chatting, or distilling into portable skills.

The interesting bit

Instead of traditional RAG, which rediscovers knowledge from scratch per query, OpenKB treats knowledge as a compounding asset: each new document updates existing concept pages and flags contradictions. It also exports subsets of this wiki as Anthropic Skills—portable folders that Claude Code, Cursor, and other agents can install natively.

Key highlights

• Vectorless retrieval for long documents via PageIndex tree structures
• Persistent wiki with automatic cross-references, entity extraction, and contradiction detection
• Multi-LLM support through LiteLLM (OpenAI, Claude, Gemini, etc.)
• Skill Factory generates redistributable agent skills with validation, evaluation, and rollback history
• Interactive chat with slash commands and session management

Caveats

• Recompilation overwrites manual edits to concept and summary pages
• Some features like lint --fix are commented out in the README as not yet available
• The entities/ layer may need explicit backfilling on older indexed documents

Verdict

Researchers and developers drowning in PDFs who want accumulated, cross-referenced knowledge should try this; those looking for a simple drop-in vector search replacement may find the compilation step heavy.