ComponentsBuild with this
rag-vectorlessemerging
PageIndex RAG
No vector DB. LLM navigates a page tree to find answers.
Overview
Eliminates the vector database entirely. Builds a hierarchical index of document summaries and uses an LLM to navigate the tree — reading page summaries, deciding which branches to explore, and synthesizing final answers. Perfect for prototypes and small corpora.
Architecture
Interactive walkthrough01
Indexing
Document Loaderlocal files
Tokenizerlowercase + split
BM25Okapi Indextf-idf scoring
Pickle Persistenceindex.pkl on disk
02
Retrieval
Query Tokenizersame pipeline
BM25 Scoringlexical relevance
Top-k Chunksranked results
03
Generation
Prompt Templatecontext window
LLM Callany provider
01
Indexing
Document Loaderlocal files
Tokenizerlowercase + split
BM25Okapi Indextf-idf scoring
Pickle Persistenceindex.pkl on disk
02
Retrieval
Query Tokenizersame pipeline
BM25 Scoringlexical relevance
Top-k Chunksranked results
03
Generation
Prompt Templatecontext window
LLM Callany provider
Summarized pipeline view. For the full interactive, scroll-driven walkthrough with clickable stages → Pipeline detail
When to use
Use when
- corpus_size < 500 documents
- zero infrastructure preference
- prototyping / proof-of-concept
- document structure is hierarchical
Avoid when
- corpus_size > 1000 documents
- latency sensitivity
- cost sensitivity (LLM calls per navigation step)
- real-time corpus updates
Compatible frameworks
llamaindexraw python
#vectorless#no-vector-db#tree-navigation#llama-index#prototype
Ready to build with PageIndex RAG?
Walk through the wizard to generate a complete, parameterized pipeline.