PageIndex FAQs

Question 1

Does PageIndex make LLM calls on the server or require an Anthropic API key?

Accepted Answer

No, PageIndex explicitly states 'Никаких LLM-вызовов внутри сервера'. It offloads all reasoning and analysis to Claude Code (Max plan), eliminating the need for an `ANTHROPIC_API_KEY` on the PageIndex server itself.

Question 2

What is PageIndex and how does it work?

Accepted Answer

PageIndex is a self-hosted server designed for Claude Code that provides vectorless RAG (Retrieval-Augmented Generation) for PDF documents. It works by generating a hierarchical 'PageIndex tree' for PDFs (one-time OpenAI API call), then allows Claude Code to navigate this tree and extract page content locally using `pdftotext` for all reasoning and analysis.

Question 3

What are the prerequisites to set up PageIndex?

Accepted Answer

To run PageIndex, you'll need Node.js (≥18), a local Python environment with the PageIndex repository cloned and its requirements installed, and `poppler-utils` for extracting text from PDFs. An OpenAI API key is also required for the initial document indexing process.

Question 4

Can PageIndex be used with OpenRouter instead of direct OpenAI API?

Accepted Answer

Yes, PageIndex supports OpenRouter. You can configure it by setting `OPENAI_BASE_URL` to `https://openrouter.ai/api/v1`, providing your OpenRouter API key, and specifying an appropriate model like `openai/gpt-4o` for tree generation.

Question 5

What are the main costs associated with using PageIndex?

Accepted Answer

The primary cost is a one-time OpenAI API call for generating the PageIndex tree for each document, estimated at ~$0.10–0.50 per document. Document searching and analysis are then free, leveraging your existing Claude Code Max plan. Storage is handled locally.

PageIndex

PageIndex

主要功能

使用案例

主要功能

使用案例