Vectorless RAG for long, complex documents — accurate, traceable, context-aware, and cost efficient.
<cite doc="report.pdf" page="43"/>pip install -U pageindexAny model works — OpenAI here as an example.
export OPENAI_API_KEY="key"Generate a tree index, then let your LLM or agent search it for answers.
from pageindex import PageIndexClient client = PageIndexClient( index="cloud", # index and store in PageIndex Cloud chat="gpt-5.6-sol", # your own LLM answers) doc_id = client.submit_document("report.pdf", wait=True)["doc_id"] messages = "What benchmarks are used?" for chunk in client.chat(messages, doc_id=doc_id, stream=True): print(chunk, end="", flush=True) With PageIndex, you can switch to a better model or raise the thinking effort for better accuracy instantly — no re-indexing, no re-embedding.
Average cost per question · log scale
Text lookup questions from MMLongBench-Doc-V2
See the benchmarkPageIndex gives LLMs a persistent structure to retrieve only the information they need.
Navigate the document structure and search by node relevance.
Relevant pages, with their context
Break the document into chunks and search by vector similarity.
Top-K chunks
PageIndex
98.7%
Uses reasoning-based tree search to determine relevance.
Vector DB
50.0%
Approximate search via semantic similarity.
The same SDK runs on your laptop, on our cloud, or inside your own private cloud.
Open source, on your machine, with your own model key.
Managed indexing, storage, and retrieval. Nothing to run.
All of Cloud, deployed in your own private cloud.