Metavolve-Labs/alexandria-aeternum-1k

Alexandria Aeternum — Genesis

Your Entry Point to Cognitive Nutrition

1,000 curated paintings · Masters only · 4,000+ tokens each · Free sample

Not scraped. Not auto-captioned. Translated from human knowledge.

Monet, Van Gogh, Rembrandt, Degas, Hokusai, Cezanne, and 100+ master artists.

---

Full 10K Dataset · MCP Access (2M+ Artworks) · Research Paper · Explore Full Archive · Scale With Us

---

MCP Access — AI Agent Marketplace

The complete high-resolution Alexandria Aeternum Genesis 10K — plus 2M+ artworks across 7 world-class museums — is available for AI agents via MCP (Model Context Protocol). No API keys. No subscriptions. Agents pay per-request with USDC micropayments.

Connect any MCP-compatible client (Claude Desktop, Cursor, Windsurf, custom agents):

{
  "mcpServers": {
    "intelligence-aeternum": {
      "url": "https://data-portal-172867820131.us-west1.run.app/mcp",
      "transport": "streamable-http"
    }
  }
}

13 MCP Tools Available

Tool	Paid?	Description
search_alexandria	Free	Search 2M+ museum artworks across 7 institutions
get_curated_metadata	Free (5/day)	Human_Standard metadata + image (500-1,200 tokens)
get_oracle_metadata	$0.16 USDC	Hybrid_Premium 111-field VLM deep analysis + image
get_compliance_manifest	Free	AB 2013 + EU AI Act Article 53 compliance manifests
search_datasets	Free	Browse 7 museum dataset catalogs
preview_dataset	Free	Sample images from a dataset
get_pricing	Free	Calculate pricing with volume discounts
get_agent_guide	Free	Complete API workflow documentation
list_enrichment_tiers	Free	Available enrichment tiers and pricing
get_enrichment_status	Free	Poll enrichment job status
purchase_dataset	Paid	Initiate batch dataset purchase
enrich_agent_image	From $0.16	Submit YOUR image for Golden Codex enrichment
deliver_artifacts	From $0.04	On-demand artifact delivery

Payment: x402 USDC on Base L2

No API keys. No subscriptions. AI agents pay autonomously with USDC micropayments on Base L2 (chain 8453). Call a paid tool, receive HTTP 402 with payment envelope, settle on-chain, re-call with proof.

Genesis Epoch (NOW): 20% off all prices for 90 days (started Feb 23, 2026).

Volume discounts auto-apply per wallet: 100+ records 25% off, 500+ 37% off, 2000+ 50% off.

Also Available: MonetizedMCP Broker

{
  "mcpServers": {
    "fluora-mcp": {
      "url": "https://fluora-mcp-172867820131.us-west1.run.app/mcp",
      "transport": "streamable-http"
    }
  }
}

3 tools for payment-first workflows: price-listing, payment-methods, make-purchase.

Discovery: OpenAPI Docs · llms.txt · .well-known/mcp.json · GitHub

---

What This Is

This is the free 1,000-artifact sample of the Alexandria Aeternum collection — your proof of concept for Cognitive Nutrition.

Every artifact carries 4,000+ tokens of structured semantic metadata: composition, emotion, symbolism, cultural context, and the artwork's own voice. Not scraped. Not auto-captioned. Translated from museum curatorial records through multi-model Oracle Synthesis.

Ready to go deeper? The comprehensive Alexandria Aeternum 10K delivers 10,097 artifacts with v2.0 enrichment, 29 structured columns, and production-grade CORE schema. Or access the full 10K in high resolution — plus 2M more artworks — via our MCP endpoint.

This Genesis sample uses v1.1-GCS-HYBRID schema. The 10K collection has been fully reprocessed on the production v1.1-GCS-CORE schema with additional fields (artistic_intent, composition, accession_number, schema_version). If you're starting new research, we recommend the 10K.

---

Why This Dataset Exists

Foundation models are consuming the internet. And the internet is running out.

Epoch AI estimates high-quality, non-duplicated training data will be exhausted by 2026-2028. Meanwhile, models trained recursively on their own outputs are collapsing — distributional tails vanish, variance compresses, and six-fingered hands become anatomical norms.

The bottleneck isn't compute or architecture. It's the quality, provenance, and semantic density of the training signal itself.

Standard art datasets give models semantic poverty:

Dataset	Typical Caption	Tokens/Image
LAION-Art	"painting of a woman"	~10
WikiArt	"oil on canvas"	~8
CC-12M	"old artwork"	~5
Web scrape	"untitled.jpg"	~1

Alexandria Aeternum:

Our Dataset	Rich Semantic Metadata	Tokens/Image
Golden Codex	Composition, emotion, symbolism, cultural context, artistic voice	4,000+

That's 400x the semantic density. Structured. Searchable. Trainable.

This is Cognitive Nutrition — the measurable bundle of properties (provenance integrity, semantic density, affective grounding, rights computability) that training data requires to produce robust, grounded models.

---

The Translation Paradigm

This is not AI-generated metadata. It is machine-articulated human knowledge.

Our enrichment pipeline takes museum curatorial records (human ground truth), combines them with the model's trained domain knowledge and deep visual analysis, and produces structured translations that render explicit what trained art historians recognize implicitly.

When a museum record says "oil on canvas, dark background" and our system outputs "chiaroscuro technique influenced by Caravaggio's tenebrism" — that's not hallucination. That's thoroughness where resource constraints existed before.

This distinction matters:

• Standard auto-captioning: model generates from statistical patterns -> synthetic data
• Our Translation Paradigm: model articulates from curatorial ground truth + visual evidence + domain knowledge -> machine-assisted articulation of human knowledge

---

What Makes This Different

Every artifact contains structured metadata across multiple dimensions:

Soul Whisper

The artwork speaks in first person — its eternal voice, channeled through AI:

"I am the mirror you have traveled across centuries to find. Look into my eyes, and see not my secret, but the beginning of your own." — From the spirit of Leonardo da Vinci

Emotional Archaeology

Not just "portrait" but the psychological journey: Enigmatic Serenity, Contemplation, Quiet Knowing, Timelessness.

Visual Analysis

Composition, color harmony, technique, lighting — the vocabulary of artistic mastery, structured for machine learning.

Cultural Context

Historical period, artistic movement, influences, cross-cultural resonances. The web of meaning that connects each work to human civilization.

Provenance & Rights

Every artifact includes full provenance tracking, source institution, and clear rights declarations for both the public domain original and the enriched metadata.

---

The Difference in Practice

Standard Dataset:

"portrait painting"

Golden Codex (Schema v1.1-GCS-HYBRID):

{
  "schemaVersion": "1.1-GCS-HYBRID",
  "title": "Mona Lisa (La Gioconda)",
  "creator": "Leonardo da Vinci",
  "description": "Leonardo's revolutionary techniques fused to transcend mere portraiture...",
  "visual_analysis": {
    "composition": "Stable pyramidal composition, folded arms forming the base...",
    "color_palette": {
      "dominant_colors": [
        {"color_name": "Umber Brown", "hex_code": "#635147"},
        {"color_name": "Warm Ochre", "hex_code": "#C4A35A"}
      ]
    },
    "style_and_technique": "The definitive demonstration of sfumato..."
  },
  "emotional_and_thematic_journey": {
    "primary_emotion": "Enigmatic Serenity",
    "secondary_emotions": ["Contemplation", "Mystery", "Timelessness"],
    "narrative_arc": "The viewer is drawn into a silent, intimate dialogue..."
  },
  "poetic_interpretation": {
    "emotional_archaeology": "Beneath the placid surface pulses a quiet knowing...",
    "eternal_voice": "I am the mirror you have traveled across centuries to find."
  },
  "soulWhisper": {
    "enabled": true,
    "message": "This relic is more than an image; it is a mirror of enduring wonder...",
    "artistSignature": "Curated by Metavolve Labs | Golden Codex Standard v1.0"
  },
  "provenance_and_lineage": {
    "source_institution": "Metropolitan Museum of Art",
    "provenance": "Public Domain - Preserved from museum digital collection"
  }
}

---

Three Ways to Use This Data

Tier 1: Browse & Discover (Parquet)

Best for: Quick exploration, Dataset Viewer, pandas analysis

The Parquet file (data/train.parquet) loads instantly in the HuggingFace Dataset Viewer. Scroll through 1,000 masterworks, see thumbnails, read Soul Whispers — no download required.

from datasets import load_dataset
dataset = load_dataset("Metavolve-Labs/alexandria-aeternum-genesis")

# Browse in pandas
import pandas as pd
df = dataset['train'].to_pandas()
df[['artifact_id', 'title', 'creator', 'soul_whisper']].head()

Key columns: artifact_id, title, creator, description, soul_whisper, image_url, soulmark_hash

---

Tier 2: Training Ready (WebDataset)

Best for: PyTorch DataLoader, LoRA fine-tuning, batch processing

The WebDataset tar (training/genesis_1k.tar) pairs each image with its complete Golden Codex JSON — the full 4,000+ token semantic package, not flattened.

import webdataset as wds

# Stream directly
dataset = wds.WebDataset("training/genesis_1k.tar")
dataset = dataset.decode("pil").to_tuple("jpg", "json")

for image, metadata in dataset:
    caption = metadata['caption']
    soul_whisper = metadata['soul_whisper']
    full_codex = metadata['golden_codex']  # Complete nested structure
    # ... your training loop

---

Tier 3: MCP Agent Access (Live API)

Best for: AI agents, programmatic access, real-time enrichment, production pipelines

Access the complete high-resolution Genesis 10K and 2M+ artworks from 7 museums via MCP. Your AI agent searches, discovers, and purchases artifacts autonomously.

# Example: AI agent workflow via MCP
# 1. Search for Impressionist paintings
# search_alexandria(query="monet impressionist water lilies", limit=20)

# 2. Preview metadata (free, 5/day)
# get_curated_metadata(artifact_id="met_437133")

# 3. Get full 111-field Golden Codex analysis ($0.16 USDC)
# get_oracle_metadata(artifact_id="met_437133")

# 4. Submit your own image for enrichment
# enrich_agent_image(image_url="https://...", tier="hybrid_premium")

Connect your agent now

---

Tier 4: Provenance Chain (Infused + Registry)

Best for: Enterprise, compliance, self-contained artifacts

Infused PNGs carry their metadata inside the file via XMP. Extract anywhere, verify integrity, prove provenance.

# Preview samples included
samples/infused_preview/
  GENESIS-met_436535.png   # XMP metadata embedded
  GENESIS-met_437133.png
  manifest.json            # Soulmark hashes for verification

# Extract metadata with ExifTool
exiftool -XMP:all GENESIS-met_436535.png

Full collection (10,000+ infused artifacts): Available via the 10K comprehensive dataset and Golden Codex Registry API.

---

Research Foundation

This dataset is the sample release from our published research program on data quality for foundation models.

The Papers

#	Paper	DOI	Key Finding
1	The Density Imperative	10.5281/zenodo.18667735	Dense metadata improves VLM capability by +25.5%; sparse captions degrade it by -54.4%
2	Cognitive Nutrition for Foundation Models	10.5281/zenodo.18667742	Production validation: 10,097 artifacts, 181.3 min, 100% success rate
3	The Entropy of Recursion	10.5281/zenodo.18436975	PEST framework + Cognitive Nutrition hypothesis
4	Perceptual Compute Offloading	10.5281/zenodo.18667749	PCO architecture for robotics perception

Key Concepts

Cognitive Nutrition (CN): The measurable bundle of properties that training data requires to produce robust, grounded models — provenance integrity, semantic density, affective grounding, rights computability, and high-entropy distributional coverage.

PEST Framework: Provenance-Embedded Semantic Transport. A four-layer metadata ontology (Provenance, Semantic Topology, Affective, Rights) anchored by a Persistent Provenance Pointer (PPP) that survives the "Social Media Gauntlet." Validated at 100% recovery across four transformation classes.

Multi-Model Oracle Synthesis: The enrichment pipeline orchestrates complementary frontier models under human-authored prompt frameworks refined through thousands of iterations. This produces enrichment that no single model achieves alone.

Triple Hash Integrity Protocol: Every artifact carries three cryptographic fingerprints:

Hash	Algorithm	Purpose	Tolerance
Pixel Integrity	SHA-256 of pixel data	Detects any alteration	Zero (exact match)
Perceptual (PPP)	pHash 256-bit	Survives compression/resize	Up to 80% degradation
Soulmark	SHA-256 of canonical JSON	Detects metadata tampering	Zero (exact match)

The Amendability Framework: A proposed living curation model where originating institutions retain governance over their enriched collections. Supports versioned corrections, human verification tagging, and trust-level field filtering.

Citation

@article{macpherson2026density,
  author = {MacPherson, Tad},
  title = {The Density Imperative: How Metadata Richness Shapes Vision-Language
           Model Performance in Cultural Heritage AI},
  year = {2026},
  doi = {10.5281/zenodo.18667735},
  institution = {Metavolve Labs, Inc.}
}

@article{macpherson2026cognitive,
  author = {MacPherson, Tad},
  title = {Cognitive Nutrition for Foundation Models: Empirical Validation of the
           PEST Framework for Provenance-Verified, Semantically Dense Training Data},
  year = {2026},
  doi = {10.5281/zenodo.18667742},
  institution = {Metavolve Labs, Inc.},
  note = {Companion dataset: Alexandria Aeternum 10K}
}

@article{macpherson2025entropy,
  author = {MacPherson, Tad},
  title = {The Entropy of Recursion: Provenance-Embedded Semantic Transport
           as a Countermeasure to Model Collapse and Context Blindness},
  year = {2025},
  doi = {10.5281/zenodo.18436975},
  institution = {Metavolve Labs, Inc.}
}

@dataset{alexandria_aeternum_genesis_2026,
  author = {Metavolve Labs, Inc.},
  title = {Alexandria Aeternum Genesis: 1K Sample with Soulprint Enrichment},
  year = {2026},
  publisher = {Hugging Face},
  doi = {10.5281/zenodo.18359131},
  version = {1.0},
  url = {https://huggingface.co/datasets/Metavolve-Labs/alexandria-aeternum-genesis}
}

---

Museums in the Alexandria Aeternum Collection

Access all 7 institutions via MCP:

Institution	Artworks	License
Metropolitan Museum of Art	375,000	CC0
Rijksmuseum	709,000	CC0
Smithsonian Institution	185,000	CC0
National Gallery of Art	130,000	CC0
Art Institute of Chicago	120,000	CC0
Cleveland Museum of Art	61,000	CC0
Paris Collections (Louvre, Orsay, Rodin)	45,000	CC0

Total: 1,625,000+ artworks searchable now. This Genesis sample contains 1,000 curated paintings from the Metropolitan Museum of Art.

---

Dataset Statistics (This Release)

Metric	Value
Collection Size	1,000 curated paintings
Curation	Paintings only — no drawings, sculptures, or decorative arts
Master Artists	668 works from Monet, Van Gogh, Rembrandt, Degas, Hokusai, Cezanne, Renoir, Rubens, and 100+ masters
Schema Version	v1.1-GCS-HYBRID
Enrichment Version	1.0
Avg Tokens/Image	4,200
Source	Metropolitan Museum of Art Open Access
Time Span	3000 BCE - 1950 CE
License	CC-BY-4.0

---

Scale With Us

This free sample is your proof of concept. When you're ready to scale:

Tier	Artifacts	Schema	Access
Free (This Release)	1,000	v1.1-GCS-HYBRID	Download now
Comprehensive	10,097	v1.1-GCS-CORE v2.0	alexandria-aeternum-10k
MCP On-Demand	2,000,000+	111-field Golden Codex	Connect via MCP — pay per artifact
Professional	100,000	v1.1-GCS-CORE	Contact us
Enterprise	1M - 50M+	v1.1-GCS-CORE	Custom partnership

MCP Pricing (Genesis Epoch — 20% Off)

Product	Price	Notes
Search	Free	Unlimited queries across 2M+ artworks
Curated Metadata	Free (5/day)	Human_Standard + image
Oracle Metadata	$0.16 USDC	Full 111-field Golden Codex + image
Enrichment-as-a-Service	From $0.16 USDC	Submit YOUR image for Golden Codex analysis
Batch Purchase	Custom	Volume discounts auto-apply

Volume Discounts: 100+ records 25% off · 500+ 37% off · 2000+ 50% off

---

Pipeline Tools (Available via MCP)

Tool	Description	Price
SD 3.5 Large + T5-XXL	Image generation on NVIDIA L4 GPU with LoRA support	TBD
ESRGAN x4 Upscaler	Real-ESRGAN super-resolution (1024px to 4096px in ~1.15s)	$0.10
Nova (Metadata Creation)	111-field Golden Codex VLM analysis via Gemini	$0.20
Atlas (Metadata Infusion)	XMP/IPTC/C2PA embedding + SHA-256 Soulmark + hash registry	$0.10
Aegis (Verification)	"Shazam for Art" — perceptual hash provenance lookup	Free
Archivus (Arweave Storage)	Permanent 200+ year storage — pay USDC, no AR needed	TBD
Mintra (NFT Minting)	Polygon NFT with full Golden Codex metadata on-chain	TBD

---

Pioneer With Us

For Researchers

• Test the Cognitive Nutrition hypothesis — Does semantic density measurably improve model robustness?
• Benchmark against LAION/WikiArt — Controlled ablations with equivalent compute
• Validate the Translation Paradigm — Is machine-articulated human knowledge more valuable than synthetic captions?
• Extend the schema — Propose new metadata dimensions

For Builders

• Fine-tune with depth — LoRAs that capture artistic meaning, not just visual patterns
• Build semantic search — Find art by emotion, symbolism, or cultural context
• Test provenance systems — Triple hash + LSH registry as a reference implementation
• Connect your AI agent — MCP endpoint for real-time access to 2M+ artworks

For Institutions

• Curate your collections — The Amendability Framework gives you governance over your enriched data
• Join the data stewardship movement — Active institutional participation in AI training quality
• Preserve culture computationally — A library that cannot burn

---

Research Applications

Model Collapse Prevention

Train on Cognitive Nutrition vs. standard datasets. Measure distributional drift. Does dense semantic grounding prevent tail erosion?

Dense Image Captioning

Ground truth for models that describe art with scholarly depth. Benchmark your captioning systems against 4,000+ token references.

Multimodal Retrieval

Build search systems that find art by emotion, symbolism, or cultural context. Test new retrieval architectures.

Cultural AI Safety

Study how AI interprets human cultural production. What biases emerge? What gets lost? What gets preserved?

Emergent Capabilities

We genuinely don't know what happens when you train on 400x semantic density. Find out with us.

---

Schema Reference (v1.1-GCS-HYBRID)

Core Identity

schemaVersion                    - "1.1-GCS-HYBRID"
title                            - Artwork title
creator                          - Artist name
artifactId                       - Unique identifier (GENESIS-XXX)
description                      - Rich semantic analysis

Visual & Emotional Analysis

visual_analysis.composition              - Spatial arrangement
visual_analysis.color_palette.dominant_colors[]  - [{color_name, hex_code}]
visual_analysis.style_and_technique      - Artistic methods
emotional_and_thematic_journey.primary_emotion   - Core emotional response
emotional_and_thematic_journey.secondary_emotions[]  - Supporting emotions
emotional_and_thematic_journey.narrative_arc     - Viewer transformation

Poetic Interpretation

poetic_interpretation.emotional_archaeology  - Psychological depth
poetic_interpretation.eternal_voice          - Artwork's message
soulWhisper.enabled                          - Boolean
soulWhisper.message                          - First-person voice
soulWhisper.artistSignature                  - Attribution line

Provenance

provenance_and_lineage.source_institution   - Metropolitan Museum of Art
provenance_and_lineage.creation_date        - Date or period
ownership_and_rights.original_work.copyright_holder  - Public Domain
creation_credits.enrichment_system          - Golden Codex Pipeline

---

License

Component	License	Notes
Original Images	Public Domain	Metropolitan Museum of Art Open Access
This 1K Sample	CC-BY-4.0	Free for any use — research, commercial, derivatives
Attribution	Required	Cite "Metavolve Labs, Inc." and link to iaeternum.ai
Scale Beyond 1K	Contact Required	[email protected]

---

Connect

• iAeternum.ai — Full archive & MCP access
• Golden Codex — Provenance platform
• Metavolve Labs — Infrastructure for cultural permanence
• GitHub — MCP server source code
• Contact — Research partnerships & custom builds

---

The Library of Alexandria burned once.

We're building the one that can't.

---

This is Cognitive Nutrition for AI.

Provenance-verified. Semantically dense. Human-grounded.

Now accessible to AI agents via MCP.

---

Download the sample · Connect via MCP · Read the research

---

Enriched with multi-model Oracle Synthesis by Metavolve Labs, Inc.

Intelligence Aeternum