Releases·Tool Release
Vectify AI open-sources PageIndex Flash, a fast local tree-indexing engine for text-based PDFs in its SDK
PageIndex announced PageIndex Flash, an open-source tree-indexing engine for long, text-based PDFs that runs entirely on the user's own machine. It builds a hierarchical tree index from a PDF's own layout instead of having a vision model infer the outline, making indexing fast, cheap, and predictable. It is available now in the PageIndex SDK.
2 engagement2 sourcesFri, Sep 4, 2026, 04:06 PM
AI Overview
- PageIndex Flash builds hierarchical tree indexes from PDF layout on local machines, fully open-sourced and the default local-mode indexer in the PageIndex SDK
- Indexing with cheap gpt-5.6-luna costs ~$0.001 per page; 9–1,098-page documents index in ~13 seconds to 4.5 minutes
- The 62-question, 34-PDF OSS Benchmark (drawn from MMLongBench-Doc-V2) shows native PDF input costing 2.1× more at 52 pages and 16.6× more at 420 pages
- Flash handles text-based PDFs only; scanned or figure-heavy files require PageIndex Cloud's vision model, which also cites to line rather than page
Related
Social·Community
Beyond Vector RAG: How PageIndex Navigates Long Documents
Beyond Vector RAG: How PageIndex Navigates Long Documents
Beyond Vector RAG: How PageIndex Navigates Long Documents
0 engagement1 sourcerssThu, Sep 3, 2026, 04:16 AM
Products·launched
DocSlicer: fast local document parser & chunker for RAG pipelines
DocSlicer is a lightweight, deterministic document parser and chunker for local agent/RAG pipelines. It converts PDF, DOCX, PPTX, and HTML into chunks, structured tables, and a heading hierarchy, processing 100-page files in about 2 seconds without models or GPU. It solves the parsing bottleneck for developers handling large financial and legal documents.
DocSlicer is a lightweight, deterministic document parser and chunker for local agent/RAG pipelines. It converts PDF, DOCX, PPTX, and HTML into chunks, structured tables, and a heading hierarchy, processing 100-page files in about 2 seconds without models or GPU. It solves the parsing bottleneck for developers handling large financial and legal documents.
1 engagement1 sourceredditTue, Jul 28, 2026, 05:30 PM
Social·Community
Developer seeks production-ready self-hosted RAG for scientific PDFs
A developer building a RAG system for scientific papers asks the community for a production-ready, open-source, self-hosted solution rather than building from scratch. They want mature projects with high-quality PDF parsing, scientific document support, incremental indexing, hybrid retrieval, and reranking.
A developer building a RAG system for scientific papers asks the community for a production-ready, open-source, self-hosted solution rather than building from scratch. They want mature projects with high-quality PDF parsing, scientific document support, incremental indexing, hybrid retrieval, and reranking.
48 engagement1 sourceredditTue, Jul 28, 2026, 05:22 PM
Products·launched
Offline PDF chat with RAG using local LLMs and vector search
A production-oriented Retrieval-Augmented Generation (RAG) system for chatting with PDFs. It uses local LLMs via Ollama, ChromaDB for vector search, and LangChain to provide grounded answers from uploaded documents, fully offline. Aimed at users needing private, local document Q&A.
A production-oriented Retrieval-Augmented Generation (RAG) system for chatting with PDFs. It uses local LLMs via Ollama, ChromaDB for vector search, and LangChain to provide grounded answers from uploaded documents, fully offline. Aimed at users needing private, local document Q&A.
1 engagement1 sourcegithubSat, Jul 11, 2026, 03:21 AM
Products·launched
OKF-RAG-Engine: structure-aware hybrid RAG retrieval engine
OKF-RAG-Engine is an open-source RAG engine that improves retrieval by using structure-aware document processing and section-aware chunking to preserve context. It combines hybrid search with pgvector (HNSW) and BM25 to reduce retrieval failures, targeting developers building RAG applications on complex documents.
OKF-RAG-Engine is an open-source RAG engine that improves retrieval by using structure-aware document processing and section-aware chunking to preserve context. It combines hybrid search with pgvector (HNSW) and BM25 to reduce retrieval failures, targeting developers building RAG applications on complex documents.
1 engagement1 sourceredditWed, Aug 5, 2026, 12:57 PM