vectifyai/pageindex v0.4.4 was released on GitHub on October 19, 2026. This release provides a unified software development toolkit for retrieval-augmented generation tools that rely on document structure instead of a vector index. It covers both indexing and conversation, locally or in the cloud, and bundling the two tasks into a single toolkit is what this release is about.
vectifyai/pageindex is presented as a document index for vector-free, reasoning-based RAG. Retrieval-augmented generation (RAG) is a way of producing answers by handing the model the evidence it has retrieved. Rather than converting documents into vectors and matching by similarity, this project builds a tree from document layout statistics and searches along that structure. A language model is used to write the node summaries of that tree. Following document structure instead of comparing whole documents as vectors is the central choice of this project. Running it locally needs no server, no vector database, and no PageIndex API key, and the indexing pipeline can be set up without a separate service. The star count aggregated on GitHub is 39406.
However, the provided release description contains no figures for accuracy, processing speed, or cost, and presents no performance comparison. Which language model is used for the summaries, and which tools go into the toolkit, cannot be known from this description alone. How much a structure that uses a model for node summaries changes retrieval accuracy, and how far a tree that depends on document layout statistics can hold up, are too early to judge without separately verified material.