Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
/**
* The studio's one public contract: it hands a finished avatar to whoever
* embedded it. This exercises the real producer (getGLBBlobData over a real
* three.js skinned mesh) and the real postMessage envelope the host SDK
* listens for (avatar-sdk/src/creator.js).
*/
import { describe, it, expect } from 'vitest'
import * as THREE from 'three'
import { getGLBBlobData } from '../../src/library/download-utils'
import {
EXPORT_MESSAGE_SOURCE,
buildExportMessage,
isEmbedded,
postAvatarToHost,
} from '../../src/library/embed-export'
/** Minimal but real rigged avatar: one skinned quad bound to a two-bone chain. */
const buildRiggedModel = () => {
const hips = new THREE.Bone()
hips.name = 'hips'
const spine = new THREE.Bone()
spine.name = 'spine'
spine.position.y = 1
hips.add(spine)
const geometry = new THREE.PlaneGeometry(1, 2, 1, 1)
const vertexCount = geometry.attributes.position.count
const skinIndices = []
const skinWeights = []
for (let i = 0; i < vertexCount; i++) {
const y = geometry.attributes.position.getY(i)
const upper = y > 0 ? 1 : 0
skinIndices.push(upper, 0, 0, 0)
skinWeights.push(1, 0, 0, 0)
}
geometry.setAttribute('skinIndex', new THREE.Uint16BufferAttribute(skinIndices, 4))
geometry.setAttribute('skinWeight', new THREE.Float32BufferAttribute(skinWeights, 4))
const mesh = new THREE.SkinnedMesh(geometry, new THREE.MeshStandardMaterial({ name: 'skin' }))
mesh.name = 'body'
const skeleton = new THREE.Skeleton([hips, spine])
mesh.add(hips)
mesh.bind(skeleton)
const model = new THREE.Group()
model.add(mesh)
return model
}
const GLTF_MAGIC = 0x46546c67 // "glTF" little-endian
/** jsdom's Blob has no arrayBuffer(), so read it the way FileReader does. */
const readArrayBuffer = (blob) =>
new Promise((resolve, reject) => {
const reader = new FileReader()
reader.onloadend = () => resolve(reader.result)
reader.onerror = () => reject(reader.error)
reader.readAsArrayBuffer(blob)
})
describe('getGLBBlobData', () => {
it('exports a rigged model as binary glTF', async () => {
const blob = await getGLBBlobData(buildRiggedModel(), { optimized: false })
expect(blob.type).toBe('model/gltf-binary')
expect(blob.size).toBeGreaterThan(0)
const buffer = await readArrayBuffer(blob)
const header = new DataView(buffer)
expect(header.getUint32(0, true)).toBe(GLTF_MAGIC)
expect(header.getUint32(4, true)).toBe(2) // glTF 2.0 container
expect(header.getUint32(8, true)).toBe(buffer.byteLength) // declared length matches
})
it('keeps the skeleton in the exported file', async () => {
const blob = await getGLBBlobData(buildRiggedModel(), { optimized: false })
const buffer = await readArrayBuffer(blob)
const jsonLength = new DataView(buffer).getUint32(12, true)
const json = JSON.parse(new TextDecoder().decode(new Uint8Array(buffer, 20, jsonLength)))
expect(json.skins).toHaveLength(1)
expect(json.skins[0].joints).toHaveLength(2)
expect(json.nodes.map((node) => node.name)).toEqual(expect.arrayContaining(['hips', 'spine']))
})
})
describe('embedded export envelope', () => {
it('reports embedding by comparing the window to its top', () => {
expect(isEmbedded()).toBe(false) // vitest runs the page as top-level
})
it('names the source the host SDK filters on', () => {
expect(EXPORT_MESSAGE_SOURCE).toBe('characterstudio')
expect(buildExportMessage(new ArrayBuffer(8))).toMatchObject({
source: 'characterstudio',
type: 'export',
format: 'glb',
})
})
it('posts the exported avatar to the host window as a transferable', async () => {
const blob = await getGLBBlobData(buildRiggedModel(), { optimized: false })
const arrayBuffer = await readArrayBuffer(blob)
const glbSize = arrayBuffer.byteLength
const posted = []
const hostWindow = {
postMessage: (...args) => posted.push(args),
}
// Exactly what ExportMenu.saveToAccount runs when the studio is embedded.
postAvatarToHost(arrayBuffer, hostWindow)
expect(posted).toHaveLength(1)
const [message, targetOrigin, transfer] = posted[0]
expect(message.source).toBe('characterstudio')
expect(message.type).toBe('export')
expect(message.format).toBe('glb')
expect(message.glb).toBeInstanceOf(ArrayBuffer)
expect(new DataView(message.glb).getUint32(0, true)).toBe(GLTF_MAGIC)
expect(targetOrigin).toBe('*')
expect(transfer).toEqual([message.glb])
// The SDK turns the payload straight back into a model/gltf-binary Blob.
const roundTripped = new Blob([message.glb], { type: 'model/gltf-binary' })
expect(roundTripped.size).toBe(glbSize)
expect(roundTripped.type).toBe('model/gltf-binary')
})
})