PageIndex

Overview

PageIndex is a vectorless, reasoning-based RAG engine for long and complex documents. Instead of a vector index and semantic similarity search, it builds a hierarchical tree index for each document and lets an LLM reason through the tree to retrieve relevant, traceable, and context-aware information without chunking or a vector database.

Key Features

• Vectorless, reasoning-based retrieval that replaces semantic similarity search with LLM reasoning over a tree index.
• Hierarchical tree index generated per document, designed to mirror how humans navigate long reports.
• No vector database and no chunking required.
• Traceable and context-aware retrieval that can use conversation history and domain knowledge.
• Local and cloud modes: local mode handles text-based PDFs, while cloud mode adds managed indexing, OCR, image understanding, metadata, folders, and an MCP server.
• Reported state-of-the-art 98.7% accuracy on the FinanceBench financial document QA benchmark.

Use Cases

• Question answering over financial reports and FinanceBench-style financial documents.
• Retrieval from legal documents and regulatory filings.
• Search and analysis of technical manuals, medical literature, and academic textbooks.
• Reasoning over long, complex professional documents that require contextual understanding and multi-step reasoning.

Getting Started

• Install the PageIndex SDK with pip install -U pageindex.
• Set your OpenAI API key, then create a PageIndexClient with index and chat model choices.
• Submit a document such as report.pdf to receive a doc_id, then call chat with a question and the doc_id to get an answer.
• Configure other models, streaming, multi-document search, and citations through the SDK client.

Deployment & Requirements

• Local mode requires the PageIndex Python SDK and your own LLM API key, such as an OpenAI API key.
• Cloud mode requires a PageIndex API key and provides managed parsing, OCR, image understanding, tree-index construction, and cloud storage.
• Dedicated VPC or on-premises deployment is available by contacting the vendor.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• Local operation depends on your own LLM API key, and cloud operation depends on a PageIndex API key.
• The PageIndex File System, which extends reasoning to an entire corpus, is Cloud-only.
• The open-source local setup is best suited to text-heavy PDFs and local workflows; scanned or image-rich documents may need the cloud service.