A free, open library of game-ready 3D assets for people and for the AI agents building alongside them. 22,000+ GLB models across 350+ packs and 200+ walkable demos, every one CC0 and free for commercial use, with no account needed to search or download. Uploads are validated, re-encoded and previewed automatically before a human reviews them, so anyone starting out in game development can build a scene without spending money, or LLM tokens, generating the models themselves.
Learning to build a 3D game runs into the same wall early: the models. A single character or a set of props can take days to make, and generating a whole world with an AI model burns tokens quickly, with no guarantee the files will load cleanly when they arrive.
So I built the library I wanted when starting out. Every model is a clean GLB that drops into three.js, Blender, Godot or Unity without extra decoders, everything is CC0 for personal and commercial use, and the whole catalogue is readable by a person or an agent. No account, no tier, no paywall, and every asset an agent pulls from it is one it did not spend tokens generating.
The catalogue, the pipeline behind it, and the ways in and out: browser, REST API and MCP.
Every upload follows the same path, whether it comes from the browser or an agent: validation, quantized geometry, WebP textures, rendered thumbnails and posters, a blurhash placeholder and per-model stats. Then it goes to a human for review.
350+ kits built around a shared palette, scale and materials. Each ships a composed starter scene, a manifest of individual model URLs and a preview, so a level starts from pieces that already belong together.
200+ small environments, each built from one pack and played in the browser with WASD and the mouse, or touch on a phone. See the real scale and style of a pack before downloading any of it.
A streamable HTTP MCP server with 20 tools, a versioned REST API with OpenAPI, per-asset Markdown and llms.txt. One command connects Claude Code, Codex, Cursor or VS Code, and an agent can search, fetch, upload and check review status.
Automation does the repetitive work. A person still signs off every model that goes live.
A contributor uploads a file, or an agent submits one from a URL, and the upload is finalised through the API with a signed URL.
Validation, quantization, WebP textures, previews, blurhash and stats run automatically, in one pass, on every file.
Assets wait in a review queue. Licences and AI attribution are confirmed before anything is published.
The asset goes live at an immutable, CORS-enabled URL, filed by category, theme and style, and free to hotlink.
A top-level look at how it's put together, from the upload that starts the pipeline to the CDN URL a game loads at runtime.
A person uploads in the browser, or an agent submits a model from a URL and finalises the upload through the API. Signed upload URLs keep the storage credentials out of the client, and nothing reaches the catalogue until the file has been checked.
Geometry is quantized and textures are re-encoded to WebP and embedded, so every model is a self-contained binary glTF that three.js loads without Draco or KTX2 decoders. The same run produces thumbnails, posters, a blurhash placeholder and the stats shown on each asset page.
Processed assets queue for review. A person checks the model and its metadata, the contributor acceptance of the CC0 dedication is recorded, and any AI involvement is attributed to a named model. Versions are kept, so a correction never breaks an existing link.
Published files live at a versioned, immutable URL on the CDN, served with CORS open to every origin and no cookies or keys. A model is downloaded once and cached, rather than re-fetched on every build.
Every asset is filed by category, theme and style. Packs group models that share a palette, scale and materials, and each one ships a composed starter scene and a manifest of individual CDN URLs alongside a walkable demo.
The whole catalogue is readable through a streamable HTTP MCP server and a versioned REST API, with per-asset Markdown, llms.txt and a skill for agent clients. Reads need no account or key; writes use a bearer key tied to the contributor.
I've spent 7+ years in paid advertising, agency-side and in-house, most recently as Head of Paid Search for 3+ years at Soar With Us, a D2C agency, growing brands to 8+ figures. I've since started Pavegrowth to continue my love of building and growing brands with that expertise.
Alongside the ad accounts I build Shopify themes, iOS apps and software products, including my own. Growth marketing and development under one roof. I utilise AI workstreams to support moving faster & streamlining internal & client workflows to ensure efficiency, whilst the strategy, judgement and accountability are mine. You talk directly to the person doing the work.
Thirty minutes to get to know your brand, your goals and where the growth is. No pitch deck, just a straight conversation about where we'd take it next.
I'll get back to you shortly about your request. Looking forward to learning more about your project & goals!
