PageIndex

🚀 Description

VectifyAI/PageIndex — a vectorless, reasoning-based retrieval framework for long documents. Instead of chunking a PDF and embedding the pieces into a vector DB, PageIndex builds a hierarchical tree index (table-of-contents-like) over the document and lets the LLM reason over the tree to navigate to the right section — retrieval by structure, not by cosine similarity.

It targets the known challenge of long-document retrieval for LLMs: context limits and context rot on long PDFs. Short docs are read in full; long PDFs (≥20 pages by default) go through PageIndex tree indexing.

🧩 Features

• Vectorless — no embeddings, no vector database, no chunk-and-embed pipeline.
• Tree index — a hierarchical summary/section tree the LLM walks to find relevant context.
• Reasoning-based retrieval — the model reasons over the index rather than matching vectors, giving context-aware, explainable navigation.
• Local-first — runs locally via the open-source version with no external dependencies; PageIndex Cloud (docs.pageindex.ai) adds capabilities for large/complex PDFs.
• Ecosystem — PageIndex MCP (MCP server), ChatIndex, ConDB.

🗒️ Reasoning for

PageIndex is the retrieval engine under OpenKB — the “vectorless PageIndex” that lets OpenKB compile and query long documents without standing up a vector store. It fits the same structural-retrieval thesis as Structural Retrieval for Code: for structured content, navigating an explicit index beats blind vector search. Complements Progressive Disclosure (index-first priming) and document parsers like MinerU / Unlimited-OCR that produce the text PageIndex then organizes. Directly relevant to the LLM Wiki pattern this vault runs on.

🔗 Links

• Repo: https://github.com/VectifyAI/PageIndex
• Docs / Cloud: https://docs.pageindex.ai/
• MCP: https://github.com/VectifyAI/pageindex-mcp

🔗 Related notes

• OpenKB — compile-to-wiki CLI that uses PageIndex as its retrieval layer
• OpenWiki — sibling docs-wiki tool referencing vectorless PageIndex retrieval
• LLM Knowledge Bases — the LLM-Wiki pattern PageIndex serves
• Structural Retrieval for Code — the “structure beats vectors” thesis, for code
• Progressive Disclosure · MinerU · Unlimited-OCR — index-first retrieval and doc-parsing neighbors

---

Template: tool