vs Vector RAG
More accurate, at lower cost
Answer accuracy
Cost per 1,000 queries
The document retrieval layer for AI
Vectorless RAG for long, complex documents — accurate, traceable, context-aware, and cost efficient.
How PageIndex works
Sourcereport.pdf — a 120-page annual report
A 120-page PDF arrives — headings, sections, page numbers, and nothing a machine can navigate.
Operating margin was 18.4% in 2023.
<cite doc="report.pdf" page="43"/>Get started in seconds
0
Install the SDK
$
pip install -U pageindex1
2
Prepare your LLM model
Any model works — OpenAI here as an example.
$
export OPENAI_API_KEY="key"3
Index a document and query it
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) Retrieval accuracy scales with the model
With PageIndex, you can switch to a better model or raise the thinking effort for better accuracy instantly — no re-indexing, no re-embedding.
Retrieval accuracyThinking effort none
Average cost per question · log scale
Text lookup questions from MMLongBench-Doc-V2
See the benchmarkCompare ways to query a document
PageIndex gives LLMs a persistent structure to retrieve only the information they need.
63 questions on NVIDIA's 10-Ks, six embedding models at top-k 10 to 100. Every system answers with the same model and is graded against reference answers.
vs LLM File Input
Lower cost
PageIndex
1×
Every query only reads the pages on the path it walks.
LLM File Input
16×
Every question reads the whole document.
Cost per question on a 420-page document.
Every claim points back to a block
Not just a page number. A citation names the layout block it came from, and that block resolves to a bounding box — so your own viewer can highlight the exact region.
Answer
Asia segment revenue was $4.21B in fiscal 2024, up 12.4% year over year.
Block-level references are a PageIndex Cloud feature; local citations stop at the page.
View DocumentationRun it anywhere
The same SDK runs on your laptop, on our cloud, or inside your own private cloud.
PageIndex Local
Open source, on your machine, with your own model key.
- Documents never leave your disk
- Cost-efficient indexing with PageIndex Flash
- Text-based PDFs, read directly — no OCR
- Open source, 35k+ stars on GitHub
PageIndex Cloud
Managed indexing, storage, and retrieval. Nothing to run.
- Our best indexing model, fully managed
- Scanned files and image understanding
- Hosted storage and file management
- Free to start — 30k+ developers already building
PageIndex Enterprise
All of Cloud, deployed in your own private cloud.
- Runs in your VPC on AWS, Azure, or GCP
- Documents and indexes stay in your cloud
- customized integrated options
- Audit logs, dedicated support, and SLA
Simple usage billing for PageIndex Cloud#
Local stays free and open source. Start with $10 in free Cloud credits. No minimum monthly cost.
Free Starter
$10in free credits
- No payment method required
- Covers your first 1,000 pages indexed
- 1,000 active pages free, every month
- Unlimited retrieval on your active pages
- Bring your own LLM
Usage Billing Rates
IndexingCharged once, when the page is indexed.$0.01/ page
Active pagesKept indexed and ready to query, billed at month end. First 1,000 free.$0.001/ page / mo
RetrievalNo retrieval fees on active pages. You pay your LLM provider directly.Unlimited
Estimate your cost
I havepages
Indexingone-time$10.00
Active pagesfirst 1,000 pages free$0.00/ mo
Retrievalunlimited queries on active pagesUnlimited
