Pageindex
📑 PageIndex: Document Index for Vectorless, Reasoning-based RAG
Last verified:
What is Pageindex?
PageIndex is a vectorless, reasoning-based RAG (retrieval-augmented generation) engine that transforms lengthy PDF documents into a searchable tree structure—like a smart, LLM-optimized table of contents. Instead of relying on vector similarity search or arbitrary chunking, PageIndex enables LLMs to navigate documents logically through agentic reasoning, finding precise answers with traceable, explainable retrieval paths. It achieves 98.7% accuracy on FinanceBench, significantly outperforming traditional vector-based RAG systems.
The tool is built for anyone who needs to understand complex long documents: financial analysts working with annual reports, legal professionals reviewing contracts, researchers analyzing academic papers, and developers building document AI applications. PageIndex Chat provides a conversational interface for everyone to upload entire books or hundred-page reports and chat without context limits. Developers can integrate PageIndex via MCP server or REST API, while enterprises get dedicated/private deployment with auditable answers and full context traces at scale.
Key features include hierarchical tree structure generation that maintains the document's logical flow, precise page-level referencing with exact page numbers and citations for every answer, chunk-free segmentation following natural document structure, and the ability to scale to thousands of pages. The platform supports multi-document comparison to reveal trends and inconsistencies across reports, deep reasoning that interprets and connects ideas across sections, and human-in-the-loop collaboration where users can guide the AI's reasoning.
PageIndex is ideal for financial reports, legal documents, technical manuals, research papers, or any document exceeding typical LLM context limits. It requires no vector database, no embeddings, and no chunking—documents are transformed into hierarchical trees that LLMs traverse using tree search algorithms combining LLM-based and value-based reasoning for efficient, accurate retrieval.
Pageindex pricing
Pricing model: Freemium
Free Trial: 200 free credits, up to 200 active pages, basic MCP & API access, community support. PageIndex Chat Free: $0/month with 1,000 free pages, 100 free chat messages, basic MCP access, community support. Standard Developer: $30/month or $300/year (save 17%) with 1,000 monthly credits, up to 10,000 active pages, full MCP & API access, page-level references, vision understanding. Pro Developer: $50/month or $500/year (save 17%) with 2,000 monthly credits, up to 50,000 active pages, lab features. Max Developer: $100/month or $1,000/year (save 17%) with 6,000 monthly credits, up to 500,000 active pages, multiple workspaces, priority support. Chat Pro: from $20/month with full chat and MCP access, ability to purchase more pages. Team: custom pricing with dedicated infrastructure, multiple users, private deployment option. Page indexing costs 1 credit per page (one-time fee). Top-up credits cost $0.01 per credit and never expire. Enterprise: custom pricing with volume discounts, private deployment, SLAs, dedicated support.
Pageindex pros
- 98.7% accuracy on FinanceBench benchmark
- No vector database required
- No arbitrary chunking—uses natural document structure
- Exact page numbers and citations for every answer
- Handles hundreds to thousands of pages seamlessly
- Traceable and explainable retrieval process
- Hierarchical tree structure mimics human reading
- Multi-document comparison and analysis
- Deep reasoning that connects ideas across sections
- No context limits for document uploads
- Works with your own LLM (no inference billing from PageIndex)
- Open-source framework available on GitHub
- MCP server and REST API for developer integration
- Human-in-the-loop collaboration features
- 100% accuracy vs ChatGPT 5.1 Thinking's 59.1% on domain tasks
- 24.3% faster than ChatGPT 5.1 Instant on document analysis
- Vision Understanding capability for visual documents
- Top-up credits never expire and carry over
Pageindex cons
- Higher cost and latency compared to vector RAG
- Beta Chat API may have stability issues
- Free tier limited to 1,000 pages and 100 chat messages
- Requires internet connection for cloud service
- Credit-based pricing can add up for heavy usage
- Primarily optimized for PDF documents
- Learning curve for API/MCP integration
- Community support only on free/standard plans
Frequently asked questions about Pageindex
What is PageIndex and how does it work?
PageIndex is a vectorless, reasoning-based RAG framework that transforms lengthy PDF documents into a searchable tree structure—like a smart table of contents optimized for LLMs. Instead of using vector similarity search or arbitrary chunking, it enables LLMs to navigate documents logically through agentic reasoning. The system has two components: PageIndex Tree Generation creates hierarchical tree indexes maintaining the document's logical flow, and PageIndex Retrieval conducts tree search using both LLM-based and value-based algorithms to extract relevant context with precise page references.
What makes PageIndex different from traditional RAG systems?
PageIndex is completely vectorless—it doesn't use embeddings, vector databases, or arbitrary chunking like traditional RAG. Instead, it builds a hierarchical tree structure that mirrors how humans read documents, enabling LLMs to perform agentic reasoning over the structure. This produces traceable, explainable retrieval with exact page citations. PageIndex achieves 98.7% accuracy on FinanceBench, significantly outperforming GPT-4o and vector-based systems, and can distinguish between totals versus changes, read context across sections, and synthesize findings meaningfully.
What document types and lengths can PageIndex handle?
PageIndex is designed for long, complex documents including financial reports, legal documents, technical manuals, research papers, and entire books. It scales to hundreds or even thousands of pages with ease—examples include 758-page PDFs and 222-page Federal Reserve annual reports. The tool handles documents of any length without context limitations, making it ideal for any document that exceeds typical LLM context limits.
How do page-level citations work in PageIndex?
Every node in PageIndex's tree structure contains its summary and start/end page physical index, enabling pinpoint retrieval. Every answer comes with exact page numbers and citations (e.g., <doc=file.pdf;page=1>), so you can trace insights directly back to their source. The original source opens side-by-side for easy verification, building trust and making verification effortless. This transparency is a core advantage over other AI tools that don't provide precise source references.
Can I use my own LLM with PageIndex?
Yes, PageIndex offers unlimited LLM integration—you can use your own LLM via MCP or API. Your LLM provider handles inference billing, while PageIndex handles the document indexing and retrieval. This means you're not locked into a specific LLM and can choose the model that best fits your needs and budget. The PageIndex SDK is available for Python and JavaScript, making integration straightforward.
What is PageIndex Chat and how is it different from the API?
PageIndex Chat is the world's first human-like long-document AI analyst with a conversational interface available at chat.pageindex.ai. You can upload entire books or hundred-page reports and have conversations directly in your browser with no setup required. It's designed for everyone who wants to understand complex documents. The API/MCP is for developers who want to integrate PageIndex's reasoning-based retrieval into their own agents or applications. Chat provides the same core technology but with a user-friendly interface.
How does the credit pricing system work?
Page indexing costs 1 credit per page as a one-time processing fee—once indexed, pages stay available for querying anytime. Chat & Retrieval API uses token-based credits per query when using the hosted Chat API, charged based on input and output tokens. LLM integration is unlimited when you use your own LLM. Each subscription includes monthly credits that reset every billing cycle. You can top up anytime at $0.01 per credit, and top-up credits never expire and carry over across cycles, unlike monthly included credits.
Is PageIndex available as open source?
Yes, PageIndex Framework is open source and available on GitHub at github.com/VectifyAI/PageIndex/. The open-source repository allows self-hosting for those who want full control. However, the cloud service (Chat, MCP, API) is hosted by Vectify AI with freemium pricing. The open-source core enables zero cost with local models, while the cloud version has a free tier and paid plans for those who prefer managed service.
What enterprise features does PageIndex offer?
Enterprise plans include private deployment options (single-tenant, private cloud, on-premise), advanced security and compliance features, custom integrations and API access, custom pricing with volume discounts, Service Level Agreements (SLAs), and a dedicated support team. Enterprise customers get dedicated infrastructure for their team, access to the most accurate model, advanced file search, granular file access controls, support for multiple users, and the option for private deployment. Contact sales for custom pricing tailored to organizational needs.
Can PageIndex analyze multiple documents at once?
Yes, PageIndex supports multi-document comparison and analysis. You can compare and analyze multiple documents simultaneously to find patterns, trends, inconsistencies, and relationships across different reports and datasets. This capability is essential for real-world analysis that rarely stops at one source. The tool can reveal insights across reports and datasets that would be difficult to identify when examining documents individually.