All Labels

We propose a practical acquisition function for prompt/completion pairs based on the predictive entropy of the language model and a measure of certainty of the implicit preference model optimized by DPO.

We introduce Model Augmented Fine-tuning (Mafin) — a novel approach for fine-tuning a black-box embedding model by augmenting it with a trainable embedding model.

We introduce Mafin2.5, which is built based on PageIndex, with a 98.7% accuracy rate on the finance industry question-answering benchmark.

The same bet behind both PageIndex and Claude Code: skip the vector DB and let the LLM itself drive retrieval. We explore the rise of agentic retrieval over vector indexing and PageIndex's agentic, vectorless RAG framework. Where Claude Code retrieves over a codebase with simple bash tools instead of a vector database, PageIndex gives long documents an in-context tree index that an LLM agent navigates by reasoning — no chunking, embeddings, or vector store.

We argue that context blindness — the inability of vector-based retrieval to condition on full conversational and reasoning context — is a fundamental limitation of vector RAG, and outline a paradigm shift from semantic similarity to context-dependent relevance classification. In this view, retrieval becomes a relevance decision made by an LLM with full context, scaled efficiently through hierarchical tree search.

We examine the inherent limitations of OCR from an information-theoretic perspective and show why a direct, vision-based approach with PageIndex is more effective. Because flattening a 2D page into a 1D text sequence is inherently lossy, PageIndex acts as a vectorless retrieval layer that selects the relevant pages of a long document, which a VLM then reads directly as images.

We built what Andrej Karpathy described, and solved the hard part. OpenKB is an open-source CLI that compiles raw documents into a structured, interlinked wiki, powered by PageIndex for long PDFs.

PageIndex OCR is the world's first OCR model that understands documents as a whole — preserving full structure and section hierarchy across pages, instead of treating each page as an independent unit.

Experience the power of reasoning-based RAG with PageIndex Chat - our new conversational interface for intelligent document understanding.

PageIndex File System is a file-level tree layer that sits above your documents and scales the same PageIndex tree search from a single document to millions of documents in one index. It synthesizes a semantic hierarchy with virtual nodes when no usable folder structure exists, builds the tree on demand for each query, and adapts how it searches each node to stay efficient at scale.

PageIndex has been selected for GitHub's Secure Open Source Fund, supporting a broader security roadmap for long-document AI infrastructure.

PageIndex reached

PageIndex is a vectorless, reasoning-based retrieval framework that simulates how human experts extract knowledge from complex documents. Instead of relying on vector similarity search, it builds a tree-structured index from documents and enables LLMs to perform agentic reasoning over that structure for context-aware retrieval. The retrieval process is traceable and interpretable, and requires no vector DBs or chunking.

PageIndex was recognized on the Open Source Growth Index (OSSCAR) Q1 2026 by Supabase × Commit VC, ranking #14 in GitHub Star Growth and #38 Overall in the Scaling Tier.

VentureBeat covers PageIndex, the vectorless, reasoning-based RAG framework that uses tree search over document structure to reach 98.7% accuracy on FinanceBench, where vector-based retrieval typically fails.

We benchmarked PageIndex Chat against ChatGPT 5.1 on real-world long documents. PageIndex achieved 100% accuracy compared to ChatGPT 5.1's 59-82%, with faster response times and page-level traceability.

How PageIndex’s vectorless, reasoning-based RAG overcomes the challenges of traditional vector RAG in long, complex technical manuals.