aax-vision-lib 1.0.0

Universal asset memorizer: scrape URLs, memorize images/text/code/video as A-Z unlimited assets

aax 🗂️

Universal Asset Memorizer — scrape images, text, code, video, and data from any URL and store them as A–Z unlimited labeled assets.

pip install aax

---

What is aax?

aax is a Python library that memorizes everything on a URL:

Asset Type	What it captures
🖼️ IMAGE	PNG, JPEG, WebP, GIF, SVG, AVIF, BMP, TIFF …
📝 TEXT	Paragraphs, headings, lists, captions, blockquotes
💻 CODE	<code>, <pre>, inline snippets
🎥 VIDEO	<video>, iframes, embedded players
🎵 AUDIO	<audio>, podcast feeds
📊 DATA	JSON-LD, meta tags, HTML tables → structured JSON
🔗 LINK	All hyperlinks with anchor text
📄 DOC	PDF, DOCX, XLSX, PPTX linked files

Every asset gets a unique A–Z unlimited label (A, B, C … Z, AA, AB … ∞).

---

Architecture

Built on three pillars:

aax/
├── core/          ← Memorizer engine + AssetSession
│   ├── memorizer  ← scrapes URLs, extracts all assets
│   ├── session    ← A-Z labeled container for results
│   └── types      ← Asset dataclass, AssetKind enum, index_label()
│
├── vision/        ← URL Vision Checker (inspired by torchvision / vision-main)
│   └── checker    ← image size, dominant colors, webpage meta
│
├── image/         ← Image processing (inspired by image-rs / image-main)
│   └── processor  ← load/transform/save images from URL or file
│
├── data/          ← Structured data builder (inspired by serde_json / json-master)
│   └── builder    ← serialize to JSON/JSONL/CSV/SQLite, full-text search
│
└── storage/       ← Persistent A-Z vault
    └── vault      ← disk-backed long-term asset memory

---

Quick Start

1. Memorize a URL

import aax

# Scrape everything from Wikipedia's main page
session = aax.memorize("https://id.wikipedia.org/wiki/Halaman_Utama")

print(session.summary())
# ━━━ aax AssetSession ━━━
#   URL   : https://id.wikipedia.org/wiki/Halaman_Utama
#   Total : 847 assets (labeled A–ZH)
#   text  : 312
#   link  : 289
#   image : 143
#   data  : 78
#   ...

# Access by label
first_image = session["A"]      # not always image — first scraped asset
print(first_image.kind)         # AssetKind.TEXT / IMAGE / etc.

# Typed views
for img in session.images:
    print(img.label, img.src)

for text in session.texts:
    print(text.label, text.content[:80])

# Search content
results = session.search("Indonesia")
print(f"{len(results)} assets mention 'Indonesia'")

# Save everything to disk
session.save("./wiki_assets", download_images=True)

2. Check URL Vision

import aax

# What's in this URL?
v = aax.vision("https://upload.wikimedia.org/wikipedia/commons/thumb/9/9f/Flag_of_Indonesia.svg/320px-Flag_of_Indonesia.svg.png")
print(v.describe())
# [aax.vision] https://upload.wikimedia.org/...
#   Type     : image/png
#   Kind     : Image (PNG)
#   Size     : 320×213 px  (ratio 1.5023)
#   FileSize : 3.2 KB
#   Colors   : #ce1126, #ffffff, #f5f5f5, #d4d4d4, #e8e8e8

print(v.dominant_colors)   # ['#ce1126', '#ffffff', ...]
print(v.size)              # (320, 213)
print(v.is_image)          # True

# Webpage vision
vw = aax.vision("https://id.wikipedia.org/wiki/Pemerintahan_Nasional_Pertama")
print(vw.describe())
# [aax.vision] https://id.wikipedia.org/...
#   Type     : text/html
#   Kind     : Webpage (HTML)
#   Title    : Pemerintahan Nasional Pertama – Wikipedia ...
#   Desc     : Pemerintahan Nasional Pertama adalah...

3. Process Images

from aax.image import ImageProcessor

ip = ImageProcessor()

# Load from URL → transform → save
(ip.from_url("https://upload.wikimedia.org/wikipedia/commons/thumb/...")
   .resize(640, 480)
   .grayscale()
   .blur(1.5)
   .save("processed.jpg"))

# Batch download
handles = ip.batch_from_urls([
    "https://example.com/img1.jpg",
    "https://example.com/img2.png",
])
for h in handles:
    h.thumbnail(256).save(f"thumb_{h.source.split('/')[-1]}")

# Get image info
h = ip.from_url("https://...")
print(h.info())
# {'format': 'JPEG', 'size': (1920, 1080), 'mode': 'RGB', ...}

4. Build Structured Data

from aax.data import DataBuilder
import aax

session = aax.memorize("https://id.wikipedia.org/wiki/Pemerintahan_Nasional_Pertama")

db = DataBuilder()
db.ingest(session)

# Export formats
db.to_json("assets.json")                      # full JSON
db.to_jsonl("assets.jsonl")                    # one record per line
db.to_csv("texts.csv", kind="text")            # CSV of text assets
db.to_sqlite("assets.db")                      # SQLite with FTS

# Search
results = db.query("kabinet")
for r in results:
    print(r["label"], r["content_text"][:60])

# Reload from disk
db2 = DataBuilder.from_json("assets.json")
db3 = DataBuilder.from_sqlite("assets.db")

5. Persistent Vault

from aax.storage import AssetVault
import aax

vault = AssetVault("./my_vault")

# Store sessions from multiple URLs
urls = [
    "https://id.wikipedia.org/wiki/Halaman_Utama",
    "https://id.wikipedia.org/wiki/Pemerintahan_Nasional_Pertama",
]
for url in urls:
    session = aax.memorize(url)
    stored = vault.store(session)
    print(f"Stored {stored} assets from {url}")

# Retrieve
asset = vault.get("A")
content = vault.get_content("B")

# Query
images = vault.list_by_kind("image")
indonesia_assets = vault.search("Indonesia")

# Stats
print(vault.stats())
# {'total': 1694, 'labels': 'A … ZZH', 'disk_bytes': 4_200_000, ...}

---

CLI Usage

# Memorize a URL
aax memorize https://id.wikipedia.org/wiki/Halaman_Utama --out ./assets

# Only scrape images and text
aax memorize https://id.wikipedia.org/wiki/Halaman_Utama --kinds IMAGE,TEXT

# Download images too
aax memorize https://example.com --download-images

# Follow internal links (depth 2)
aax memorize https://example.com --follow-links --depth 2

# Vision check
aax vision https://example.com/image.png
aax vision https://id.wikipedia.org/wiki/Halaman_Utama --json

# Vault management
aax vault ./my_vault stats
aax vault ./my_vault list --kind image --limit 20

---

Asset Labels: A–Z Unlimited

Assets are labeled like Excel columns — never runs out:

A, B, C, … Z,
AA, AB, AC, … AZ,
BA, BB, … ZZ,
AAA, AAB, … ∞

from aax.core.types import index_label

index_label(0)    # 'A'
index_label(25)   # 'Z'
index_label(26)   # 'AA'
index_label(701)  # 'ZZ'
index_label(702)  # 'AAA'

---

Filtering Asset Kinds

from aax.core.types import AssetKind

session = aax.memorize(url, kinds=[AssetKind.IMAGE, AssetKind.TEXT])

Available kinds: IMAGE TEXT CODE VIDEO AUDIO DATA LINK DOC FONT STYLE SCRIPT ICON IFRAME UNKNOWN

---

Advanced: Multi-URL Scrape

import aax
from aax.data import DataBuilder
from aax.storage import AssetVault

urls = [
    "https://id.wikipedia.org/wiki/Halaman_Utama",
    "https://id.wikipedia.org/wiki/Pemerintahan_Nasional_Pertama",
]

vault = AssetVault("./vault")
db    = DataBuilder()

for url in urls:
    print(f"Memorizing {url}")
    session = aax.memorize(url, verbose=True)
    vault.store(session)
    db.ingest(session)
    print(session.summary())

# Export everything
db.to_sqlite("all_assets.db")
print(f"\nVault total: {len(vault)} assets")
print(vault.stats())

---

Dependencies

Core (always installed):

• requests, aiohttp, httpx — HTTP
• beautifulsoup4, lxml — HTML parsing
• Pillow — image processing
• rich, click, tqdm — CLI/output

Optional (pip install aax[vision]):

• torch, torchvision — deep vision models
• transformers — image captioning, classification
• opencv-python — advanced image ops

Full (pip install aax[full]):

• All vision deps + yt-dlp, pytesseract, pdf2image

---

Inspired By

Library	Role in aax
serde_json (json-master)	Structured data serialization, JSON A-Z asset records
image (image-main)	Image format support, decoding/encoding pipeline
torchvision (vision-main)	URL-based image loading, transform pipelines, vision checking

---

License

MIT © aax