PageIndex Agent

Vectorless, OCR-free, multimodal RAG for complex documents, with line-level citations for every answer.

Get API key
Annual report120 pagesBusiness overviewp. 1–24Financial resultsp. 25–78Risk factorsp. 79–120p. 25p. 26p. 27p. 28p. 29p. 30p. 31p. 32p. 33p. 34p. 35p. 36p. 37p. 38p. 39your documentsFinancial results, p. 25–78p. 341.Find relevant documents2.Find relevant sections3.Find relevant pages4.Find it on the page
Annual report120 pagesBusiness overviewp. 1–24Financial resultsp. 25–78Risk factorsp. 79–120p. 25p. 26p. 27p. 28p. 29p. 30p. 31p. 32p. 33p. 34p. 35p. 36p. 37p. 38p. 39your documentsFinancial results, p. 25–78p. 341.Find relevant documents2.Find relevant sections3.Find relevant pages4.Find it on the page

1. Find relevant documents

Organize and search your documents

Choose the organization that fits your project, collections with metadata or folders, and it finds the documents that answer each question.

Collections with metadata

Company research 3 documents
Example collection with document metadata
DocumentTypeYear
Annual report.pdfAnnual report2025
Q4 earnings.pdfEarnings2025
Investor letter.pdfLetter2025

File system with Folder

Workspace
  • Research
    • Acme
      • Annual report.pdf
      • Q4 earnings.pdf
      • Investor letter.pdf
    • Market reports

2. Find relevant sections

Understands document structure. Uses fewer tokens.

It builds a tree of each document’s sections, like a table of contents, and navigates it as a person would: straight to the sections that matter, each traced to its pages. It never reads the whole document, so it spends far fewer tokens.

Annual report.pdf120 pagesBusiness overviewp. 1–24Financial resultsp. 25–78Revenue and costsp. 26–40Operating performancep. 41–52Risk factorsp. 79–120

3. Find relevant pages

Finds the right pages. Gets answers faster.

Page Search narrows each section to the few pages that answer, reading text and images alike. Your LLM reads only those pages, not the whole section.

p. 25p. 26p. 27p. 28p. 29p. 30p. 31p. 32p. 33p. 34p. 35p. 36p. 37p. 38p. 39Financial resultsp. 25–78What drove revenue growth?only the relevant pages

4. Find it on the page

Visual understanding, with visual citations

Reads charts, tables and drawings, and outlines the evidence on the page.

MERIDIAN TOWER / STRUCTURAL DRAWINGSLevel 3 framing plan01 / Framing planABCD123B1B3B27.2 m9.0 m7.2 mFigure 1. Level 3 framing plan, grids A–D / 1–3.02 / Beam scheduleMarkSectionSpanB1W310 × 397.2 mB3W410 × 609.0 mIllustrative example12

What section is beam B3, and how long is its span?

Beam B3 spans 9.0 m, between grids B and C, and is a W410 × 60 section.

Select a citation to highlight its source.

Works with any LLM

PageIndex with different LLMs on the same questions. Pick the balance of accuracy and cost that suits you, and switch models any time.

Accuracy and cost across selected LLMs using PageIndexAn illustrative pencil chart of DeepSeek, GPT Terra, Gemini, and Claude Opus. The blue line connects the Pareto frontier among these selected configurations. Positions reflect benchmark results; numeric scales are omitted.AccuracyHigherLower costHigher costDeepSeek V4.1 Flash (max)DeepSeek V4.1 FlashGPT-6.1 Terra (max)GPT-6.1 TerraGemini 4 Argon (high)Gemini 4 ArgonClaude Opus 5.5 (max)Claude Opus 5.5a model for every trade-off
Illustrative · frontier among selected modelsExplore the benchmark

More accurate than vector RAG, at lower cost

PageIndex against six embedding models on the same questions, with the same LLM answering.

AccuracyCost per 1,000 queries40%60%80%100%$0$25$50$75$100$125top-10top-30top-50top-10092.1% · $134Voyage 4Voyage 4 LargeVoyage 4 LiteGemini Embedding 2OpenAI 3-largeOpenAI 3-smallPageIndex 100% · $110
Vector RAG at top-k 10, 30, 50 and 100Explore the benchmark

Start with $10 of free credit.

Get API keyTry the examples