Open format · v1.0

Documents that open like a PDF.
Parsed by AI in one line.

ANDF is a new document format. Every .andf file opens instantly in any browser — no install, no extension. And the full document data is embedded as JSON for AI tools to extract in a single regex.

Get started ↓ Download for macOS
localhost:51423/research_paper.andf

Attention Is All You Need

Dr. Alice Chen · Dr. Bob Martinez · 2026-03-02

The Transformer architecture, introduced in 2017, fundamentally changed the landscape of natural language processing. In this retrospective, we revisit the original design decisions and examine the emergent capabilities that large-scale transformers have displayed.

💡 Tip — For a fixed compute budget, train a smaller model on more data.
FactorExponentDescription
Parameters0.076Fixed data & compute
Tokens0.095Fixed params & compute
FLOPs0.050Optimal allocation
1
How it works

One file. Two superpowers.

An .andf file is a self-contained HTML file with structured JSON embedded inside a script tag.

1

Open in any browser

Double-click a .andf file. It opens in your browser looking exactly like a PDF — toolbar, white pages, page numbers, zoom. No install, no extension, no server.

2

Parse with one regex

AI tools extract the full structured document — sections, blocks, semantic metadata, entities — from a single <script type="application/andf+json"> tag.

3

Generate with Python

Build documents programmatically with a fluent API. Headings, tables, code blocks, callouts, images — all the block types you need.

Quick start

Build a document in minutes.

Pure Python, zero dependencies.

Python 
from andf import ANDFDocument

doc = ANDFDocument()
doc.set_metadata(
    title="Q4 2025 Report",
    authors=["Finance Team"],
)
doc.set_theme("corporate")

sec = doc.add_section("summary", "Executive Summary")
sec.paragraph("Revenue grew **23%** YoY to $4.2B.")
sec.callout("Record quarter — highest revenue ever.", variant="success")
sec.table(
    headers=["Metric", "Q4 2025", "Change"],
    rows=[["Revenue", "$4.2B", "+23%"],
          ["Margin",  "31%",   "+3pp"]],
)

doc.save("report.andf")     # → open in browser like a PDF
For AI

Every document is AI-ready.

No preprocessing. No PDF parsing libraries. One regex and you have everything.

Extract JSON (any language)
import re, json

data = json.loads(
  re.search(
    r'application/andf\+json[^>]*>(.*?)
    r'cript>',
    open("file.andf").read(),
    re.DOTALL
  ).group(1)
)

# Full document structure, instantly.
# No OCR. No layout parsing.
📄

Full text extraction

Plain text or Markdown in one call — ideal for LLM context windows.

🔍

RAG-ready chunking

Automatic chunking by section with configurable token limits.

🏷️

Semantic metadata

Every block carries role, importance score, entities, and a summary.

Features

Everything a document needs.

10 block types, 4 themes, fluent Python API.

🖥️

PDF-like browser viewer

Toolbar, page numbers, zoom 50–200%, keyboard nav, TOC sidebar, print support.

🎨

4 built-in themes

Default, Academic, Corporate, Minimal — each fully customisable.

📦

Self-contained

One file. Fonts, assets, JS all inline. Email it, share it, archive it.

🔗

References & citations

Structured reference list with DOI support and inline citation markup.

💻

Syntax highlighting

Code blocks with language detection via Prism.js — 200+ languages.

🐍

Zero dependencies

Pure Python stdlib. No Pillow, no lxml, no Pandoc. Just pip install andf.

Install

Two ways to get started.

For developers building with ANDF, or end users who just want to open .andf files.

Developers

Python library + CLI. Build, parse, and extract documents programmatically.

pip install andf
andf install  # register .andf on macOS

Requires Python 3.8+ · Zero runtime dependencies · View on GitHub

macOS users

Download the app. Double-click .andf files — they open in your browser like PDFs.

↓ Download ANDF Viewer

macOS 12+ · No Python required · 868 KB