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
/**
* Dialect-aware ERC-8004 reputation reads. Pure ethers, no DOM, so the browser
* helpers (reputation.js, the passport widget) and the server trust readers under
* api/_lib all share this one implementation.
*
* Two registry dialects exist in the wild:
*
* reference The ERC-8004 reference implementation that runs at the canonical
* 0x8004... addresses. Reputation is a per-reviewer ledger:
* getSummary(agentId, clientAddresses, tag1, tag2) averages the feedback
* of the reviewers you name and REVERTS on an empty list.
* legacy contracts/src/ReputationRegistry.sol, for instances someone deployed
* themselves: getReputation(agentId) returns (avgX100, count).
*
* The reference registry is an upgradeable proxy, so its interface is a moving part.
* Detection is by behaviour (does getClients answer?) and never by address.
*/
import { Contract } from 'ethers';
import { REPUTATION_REGISTRY_ABI, REPUTATION_REFERENCE_ABI } from './abi.js';
// getSummary walks every feedback entry of every named reviewer inside one eth_call.
// Chunking keeps an agent with thousands of reviewers under node gas caps.
const REVIEWER_CHUNK = 100;
const WAD_DECIMALS = 18n;
/** @returns {Promise<'reference'|'legacy'>} */
export async function detectReputationDialect(address, runner) {
const reference = new Contract(address, REPUTATION_REFERENCE_ABI, runner);
try {
await reference.getIdentityRegistry();
return 'reference';
} catch {
return 'legacy';
}
}
async function readReference(address, runner, agentId, { tag1 = '', tag2 = '', reviewers } = {}) {
const registry = new Contract(address, REPUTATION_REFERENCE_ABI, runner);
// ethers returns a frozen Result; copy it so it can be sliced and re-encoded.
const clients = reviewers?.length ? reviewers : Array.from(await registry.getClients(agentId));
if (clients.length === 0) return { average: 0, count: 0, reviewers: 0, dialect: 'reference' };
// Count-weighted mean across chunks, carried at 18 decimals so chunks that report
// different `summaryValueDecimals` still combine exactly.
let weightedSum = 0n;
let total = 0n;
for (let i = 0; i < clients.length; i += REVIEWER_CHUNK) {
const [count, value, decimals] = await registry.getSummary(
agentId,
clients.slice(i, i + REVIEWER_CHUNK),
tag1,
tag2,
);
if (count === 0n) continue;
weightedSum += value * 10n ** (WAD_DECIMALS - BigInt(decimals)) * count;
total += count;
}
const average = total === 0n ? 0 : Number(weightedSum / total) / 1e18;
return { average, count: Number(total), reviewers: clients.length, dialect: 'reference' };
}
async function readLegacy(address, runner, agentId) {
const registry = new Contract(address, REPUTATION_REGISTRY_ABI, runner);
const [avgX100, count] = await registry.getReputation(agentId);
const n = Number(count);
return { average: n === 0 ? 0 : Number(avgX100) / 100, count: n, reviewers: n, dialect: 'legacy' };
}
/**
* Read an agent's aggregate reputation from whichever dialect lives at `address`.
*
* `average` is in the unit reviewers submitted: three.ws star reviews are 1 to 5, and
* other ERC-8004 clients commonly submit 0 to 100. The registry does not normalise
* scales and neither does this function, so compare against a threshold in the same unit.
*
* @param {object} opts
* @param {string} opts.address Reputation registry address.
* @param {import('ethers').ContractRunner} opts.runner Provider or signer.
* @param {number|bigint|string} opts.agentId
* @param {string[]} [opts.reviewers] Count only these reviewers (reference dialect).
* Default: everyone who has left feedback.
* @param {string} [opts.tag1] ERC-8004 feedback tag filters (reference dialect).
* @param {string} [opts.tag2]
* @returns {Promise<{average:number, count:number, reviewers:number, dialect:'reference'|'legacy'}>}
*/
export async function readAgentReputation({ address, runner, agentId, reviewers, tag1, tag2 }) {
const id = BigInt(agentId);
const dialect = await detectReputationDialect(address, runner);
return dialect === 'reference'
? readReference(address, runner, id, { reviewers, tag1, tag2 })
: readLegacy(address, runner, id);
}