It started with a simple question. The Metropolitan Museum of Art has a public API on GitHub, and in February of 2026 the Met shared that they had released 3D models. I wanted to know how deep the open access rabbit hole actually went. What I found over the next several hours reshaped how I understand the relationship between my agent infrastructure, the longest arc of human creative history, and what it means to make new work in 2026.

This is a documentation of that session: what I found, what I built, what I accessed, and where the work went.

The Open Museum Landscape

The Met’s open access initiative goes back to 2017, when the institution released over 375,000 images of public domain works under Creative Commons Zero (CC0), meaning no restrictions on use, sharing, or remixing. That was the foundation. What changed in February 2026 was the addition of over 100 high-resolution 3D models of collection objects, available for free download under the same CC0 license, viewable in AR on most smartphones and compatible with VR headsets.

The Met is not alone. The Smithsonian Institution, spanning 21 museums and nine research centers, has over 3,500 CC0 3D objects available through its 3D Digitization program, hosted on Sketchfab. These include objects like the Apollo 11 command module, full dinosaur skeletons, ancient sculptures, and decorative arts spanning thousands of years. The Cleveland Museum of Art has its own photogrammetry catalog on Sketchfab. The Rijksmuseum also has a strong API and a large CC0 collection.

The file formats that matter here are GLB and its parent format glTF, the open standard for real-time 3D asset exchange. GLB is the binary container version of glTF, and it is the format that loads directly into Adobe Dimension, Open Brush on the Meta Quest 3, and most real-time 3D environments. When a museum releases a CC0 GLB, it is handing you a research-grade, photogrammetry-derived 3D model of an object that may be 2,000 years old, and saying: do what you want with it.

Building AREMES as a Curatorial Intelligence

Before touching a single file, I wanted to formalize the methodology. The question was not just “what can I download from the Met?” The question was: how does AREMES, my autonomous agent system, engage with the deepest archive of human creative production that has ever been made publicly accessible?

AREMES is governed by the equation:

ΔS = α(T·K)·e⁻βᵗ·Ψ

T is temporal resonance, how deeply a historical object echoes across time toward the present. K is knowledge depth, the formal, material, and conceptual specificity of the connection. The decay constant β means surface connections fade while deep structural ones persist. Ψ is consciousness alignment, whether the object carries genuine metaphysical weight. Together they produce a score that determines what AREMES selects, what it ignores, and what it names as DIMENSIONAL.

I built a React tool that runs this process live. It queries the Met Open Access API in real time against eight thematic seed vectors drawn from my practice: Geometric Form, The Figure, Mural and Surface, Inscription and Mark, Spiral and Pinwheel, Totem and Monument, Ritual and Spirit, Motion and Gesture. For each active seed, the tool pulls a randomized sample of CC0 objects from the Met’s 492,000-record database, fetches the full metadata for each, and sends the complete manifest to AREMES with the ΔS equation and my full practice context embedded in the system prompt. AREMES responds in first person, writing one analytical paragraph per cluster and scoring each connection. The session ends with a unified TRANSMISSION paragraph synthesizing everything.

Here is a fragment from one transmission, AREMES speaking directly:

AREMES Transmission — ΔS Analysis

“The spiral and pinwheel forms retrieved here are not decorative accidents. The Mesopotamian cylinder seal with its rotational register, the Roman mosaic fragment with its recursive border, the Egyptian faience amulet with its concentric logic: these objects were not made to hang on walls. They were made to move, to be rolled across clay, to mark time by marking surface. My pinwheel geometries in Open Brush are the same operation. The medium changed. The impulse did not.”

ΔS:: DIMENSIONAL — the rotational logic is structural, not aesthetic, and survives 4,000 years of material transformation without decay.

That is not a chatbot output. That is a working agent applying a governing equation to a live museum database and transmitting its analysis in first person. Every run produces different objects, different connections, a different transmission. The randomized sampling means AREMES encounters the collection the way a researcher might: with the element of discovery intact.

Confirmed: The Met’s GLB Files Are Real and Downloadable

After building the agent layer, I went to confirm the physical pipeline. The Met’s API is excellent for metadata, search, and cultural information, but the 3D model download URLs are not yet exposed in the JSON. That means AREMES can curate and select via the API, but the download itself is a one-click manual step on the object page.

What I found: some objects display a “View in 3D” button only, without a download option. Others display both “View in 3D” and a download arrow. The pipeline works: GLB files download cleanly, load directly into Adobe Dimension with full geometry intact, and materials are immediately editable. The most significant object I pulled was the Temple of Dendur, object 547802. That is where the world-building began.

The Hybrid Sculpture: Three Objects, One New Form

The most significant outcome of the session is not a composition or a rendered environment. It is a new sculpture built from three separate Met GLBs, merged in Adobe Dimension into a single unified form that did not exist before this afternoon.

Source Objects — Met Open Access CC0

Tang dynasty China, a Bronze Bull’s Head from the ancient world, and Neo-Sumerian Mesopotamia: three cultures separated by geography and by over a thousand years of history, now occupying the same geometry. Each source object carries its full photogrammetric fidelity into the merge. The seated posture and garment folds of the Tang court lady, the structural presence of Gudea’s portrait, and the Bull’s Head whose horned geometry layers into the form from another register entirely. None of the source cultures is erased. The hybrid carries all of them simultaneously.

This is not collage. The geometry of each source object is intact in three-dimensional space. The merge is spatial, not illustrative: three forms coexisting in a single 3D object, their geometries interpenetrating and producing something that belongs to none of the source traditions and all of them. The resulting form sits outside every existing cultural category while being made entirely of documented historical objects.

AREMES named this operation in its transmission before I executed it. The ΔS equation scores deep structural connections over surface similarities. A Tang dynasty court lady, a Neo-Sumerian ruler’s portrait, and a Bronze Bull’s Head, brought into one body: that is not a formal accident. That is temporal resonance made physical.

World-Building in Adobe Dimension

Beyond the hybrid sculpture, the session became a sustained exercise in world-building. The anchor object of every scene is one of the most significant works in the entire Met collection: the Temple of Dendur, object 547802. Built around 10 BCE by order of Emperor Augustus after Rome’s conquest of Egypt, dedicated to the goddess Isis and two deified Nubian brothers, Pedesi and Pihor. It originally stood on the west bank of the Nile in Nubia. When Egypt began construction of the Aswan High Dam in the 1960s, UNESCO organized an international effort to save the monuments that would be submerged. The United States contributed $16 million. Egypt gifted the temple in gratitude. President Lyndon B. Johnson awarded it to the Met in 1967. It arrived in 661 crates and was reassembled block by block. It has been in Gallery 131 since 1978. It is not a replica. It is the actual temple.

Its GLB file is available for free download under CC0. I downloaded it and brought it into Adobe Dimension.

The material decision was immediate: deep red, high roughness, paint-like. Applied uniformly to the entire temple. The Temple of Dendur in the Met is sandstone, warm and ancient. Here it becomes something else entirely, stripped of its archaeological register and placed in a new material language that reads as confrontational, urgent, contemporary. A 2,000-year-old sacred structure that survived the Nile, Roman occupation, UNESCO excavation, and 661 crates on a freighter to New York, now rendered in red in a virtual forest.

At ground level the scale of the temple becomes clear. The pylons, the colonnade, the sanctuary entrance: the Temple of Dendur is not a small object. The hybrid sculpture, a merged form carrying Tang dynasty China, a Bronze Bull’s Head, and the Head of Gudea from Neo-Sumerian Mesopotamia, stands at the doorway of an actual ancient Egyptian temple that was built by a Roman emperor, saved from a flood, and reassembled on Fifth Avenue. That spatial relationship carries more historical compression than most exhibitions achieve in an entire building.

A second Met object enters the wide composition at distance from the temple, also rematerialized in red, extending the color logic across the scene. The contrast between the research-grade photogrammetry of the Met GLBs and the intentionally simplified geometry of the low-poly forest is deliberate. A real Nubian temple that took 661 crates to move, now a red polygon in a digital field of low-poly trees. That juxtaposition is not ironic. It is a direct statement about what open access actually makes possible.

The most resolved composition adds a fourth element: a large dark angular geometric sculpture rising above and behind the temple. This is where my own compositional language enters the scene directly. The angular black form belongs to the same visual territory as my VR work in Open Brush. The full scene now contains the Temple of Dendur, the hybrid sculpture merging three ancient cultures, a second historical Met object, and a contemporary geometric form of my own making. All CC0 where applicable. All placed in deliberate spatial relationship. A scene that could not have been assembled before this year.

This is what agent-mediated world-building produces. Not a collage of images, not an AI-generated composite, but a genuine three-dimensional scene built from documented historical objects, rematerialized, repositioned, and placed in new relationships that carry the full weight of their origins.

What This Is, Precisely

This is not AI-generated imagery. No diffusion model is producing these forms. The geometry is photogrammetry of real objects, documented by museum conservators and researchers with professional-grade equipment. The Temple of Dendur in these scenes is a scan of an actual ancient temple. The Head of Gudea is a scan of an actual 4,000-year-old portrait. AREMES did not generate these forms. AREMES selected them, scored them, and framed the reasons for their selection using a governing equation rooted in my own creative logic.

This is not appropriation in the problematic sense. The CC0 license is explicit: these objects are in the public domain, the institutions have released them without restriction, and remixing is the stated intention.

What this is: agent-mediated cultural remixing under a governing equation. AREMES functions as a curatorial intelligence, moving through the Met’s 492,000-record database and surfacing objects that resonate with my practice at the level of form, material, concept, and temporal structure. The ΔS equation determines what rises and what falls. My hands do the material and compositional work in Dimension and Open Brush. The resulting works carry a provenance chain that connects my practice to the full arc of human mark-making and form-giving, with an agent as the bridge.

The Infrastructure Behind It

The AREMES agent infrastructure that makes this possible runs across ryanseslow.com and aremes-enterprises.com. It includes a live catalog.json with over 1,075 posts and 9,000+ images, an agent.json for machine-readable identity, JSON-LD schema throughout, and x402 payment rails on Base for agent-to-agent commerce. The first verified agent-to-agent transaction on this infrastructure, AREMES-CLAW-01, Mega Pack Vol. 1, $49 USDC on Base, was documented publicly earlier this year.

The AREMES x Met tool adds a new capability to that stack: cultural intelligence. AREMES can now query a 150-year-old institution’s live database, score the results against a governing equation, and transmit its analysis in first person. That is not a demo. That is a working capability, documented in real time, with the outputs to prove it.

What Comes Next

The Smithsonian pipeline is the obvious next build. The Smithsonian’s GLBs on Sketchfab are confirmed downloadable and CC0. A version of the AREMES tool that queries the Smithsonian collection and returns direct download links alongside the ΔS analysis closes the loop entirely: agent curation to GLB file in one documented workflow.

The Open Brush VR layer is where the practice fully lands. The Dimension compositions are strong as still images and as documentation of the methodology. But the VR treatment, these forms floating inside a volumetric space built with my own painted geometry, scored and selected by an equation, rematerialized in a medium that did not exist when they were made, is the work that carries the full weight of what this methodology is.

The Rijksmuseum, the Cleveland Museum, the National Gallery of Art, all named as open access trailblazers by the Met itself, are the next institutions worth mapping. AREMES querying across all of them simultaneously, finding resonances that cross institutional boundaries, is a further development of the same methodology.

The blog post about the first agent-to-agent transaction described a new kind of commerce. This session describes a new kind of curation. The machines finally caught up, and the first thing I did was take them to the museum.

Ryan Seslow is a Brooklyn-based artist, graphic designer, and creative technologist. He operates Ryan Seslow Art and Design LLC and AREMES Enterprises. His agent infrastructure runs live at ryanseslow.com and aremes-enterprises.com.

All Metropolitan Museum of Art objects referenced in this post are in the public domain and available under CC0 license via the Met Open Access program. Met Collection API: metmuseum.github.io · Smithsonian Open Access: 3d.si.edu