PageIndex Emerges as Vectorless RAG Alternative for AI Retrieval
Last updated Sep 28, 2026
PageIndex builds a hierarchical tree from a document's natural structure, skipping embeddings, chunking, and vector searches. An LLM then reasons through the tree using summaries and titles for precise retrieval. On FinanceBench, the Mafin 2.5 system hit 98.7% accuracy—well above vector-based tools like Perplexity at 45% or GPT-4o at 31%—while preserving context and explainability for financial filings and technical manuals. Developers praise its potential for structured docs but debate scalability, latency, and fit for broader searches.
This story is a summary of posts on X and may evolve over time. Grok can make mistakes, verify its outputs.
The entire RAG industry is about to get cooked. Researchers developed a new RAG approach that bypasses almost everything traditional RAG depends on.
- No vector DB
- No data embeddings
- No chunking
- No similarity search It's called PageIndex. Instead of splitting yourShow more
The entire RAG industry is about to get cooked. Researchers developed a new RAG approach that bypasses almost everything traditional RAG depends on.
- No vector DB
- No data embeddings
- No chunking
- No similarity search It's called PageIndex. Instead of splitting yourShow more
Obviously you can replace a reranker with Jev But I don't see how you replace embeddings (i.e. first-cut search) unless your dataset is so small that you never needed embeddings in the first place
98.7% is a result for Mafin 2.5, a full system built on PageIndex. Even its authors say FinanceBench mainly tests questions answered from a single document. Show me an ablation proving the tree index caused the gain, plus a matched comparison against strong hybrid RAG onShow more
55
1.1K1.1K
Been wanting people to use Jev for this for a while now :) Semantic hierarchical search! just like a human would. check out our hierarchical classification cookbook
The entire RAG industry is about to get cooked. Researchers developed a new RAG approach that bypasses almost everything traditional RAG depends on.
- No vector DB
- No data embeddings
- No chunking
- No similarity search It's called PageIndex. Instead of splitting yourShow more