openindex 0.1.4

Compile PDFs into a queryable wiki.

OpenIndex

Overview

OpenIndex parses PDF documents into a hierarchical section tree and compiles them into a persistent, cross-linked wiki that agents can query.

It combines two projects:

• PageIndex — LLM-based hierarchical section extraction from PDFs
• OpenKB — compiles documents into a queryable wiki with cross-document concept pages

Unlike traditional RAG (which rediscovers knowledge on every query), OpenIndex compiles once: sections are indexed, summaries generated, concept pages created with bidirectional links, and a structured wiki is written to disk. An agent can then search the wiki to answer questions precisely.

Table of Contents

• Installation
• Usage
  • Index a PDF
  • Query the wiki
• License

Installation

From PyPI:

pip install openindex

From source:

uv pip install git+https://github.com/hienhayho/openindex.git

Usage

Set environment variables (or use a .env file):

OPENAI_MODEL_NAME=...
OPENAI_BASE_URL=https://api.openai.com/v1
OPENAI_API_KEY=
OPENAI_EXTRA_BODY={}

Note: openindex works with any OpenAI-compatible API server (OpenAI, vLLM, Ollama, LM Studio, etc.). Set OPENAI_BASE_URL to point to your server.

Index a PDF

Runs the full pipeline: section extraction → verification → tree building → summaries → wiki generation.

import os
import json
from dotenv import load_dotenv
from openindex import WikiIndex, TreeConfig

load_dotenv()

index = WikiIndex(
    model_name=os.getenv("OPENAI_MODEL_NAME"),
    base_url=os.getenv("OPENAI_BASE_URL"),
    api_key=os.getenv("OPENAI_API_KEY"),
    extra_body=json.loads(os.getenv("OPENAI_EXTRA_BODY", "{}")),
    config=TreeConfig(max_parallel_llm_calls=8),
)

result = index.build_wiki_sync("paper.pdf", "./wiki")
WikiIndex.print_result(result)

See tools/index.py for a full example.

Output wiki structure:

wiki/
├── index.md              # master catalog
├── summaries/<doc>.md    # section tree with page ranges
├── concepts/<slug>.md    # cross-document concept pages
└── sources/<doc>.json    # full per-page text

Query the wiki

The query agent searches the compiled wiki to answer questions, fetching only the relevant pages.

import os
import json
from dotenv import load_dotenv
from openindex import WikiQueryAgent

load_dotenv()

agent = WikiQueryAgent(
    wiki_dir="./wiki",
    model_name=os.getenv("OPENAI_MODEL_NAME"),
    base_url=os.getenv("OPENAI_BASE_URL"),
    api_key=os.getenv("OPENAI_API_KEY"),
    extra_body=json.loads(os.getenv("OPENAI_EXTRA_BODY", "{}")),
)

answer = agent.ask_sync("What is RAG?")
print(answer)

See tools/query.py for a full example.

License

Apache 2.0. See LICENSE for details.

This project incorporates code from:

• PageIndex — MIT License
• OpenKB — Apache 2.0 License