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.
ENS / DNS Agent Claim — v1
A convention for letting a domain prove it owns a three.ws (and vice versa). Bidirectional binding between human-readable names and on-chain agent identities.
Why
ERC-8004 gives an agent a CAIP-10 reference. Humans don't memorize eip155:8453:0x8004A169…/42. Without a name binding, every "this is the official three.ws for example.com" claim is just an unverifiable assertion in a card.
This spec defines two records that are cheap to set, easy to verify, and require no new on-chain infrastructure.
Forward record (name → agent)
ENS
Set the text record agent on the ENS name:
agent = eip155:<chainId>:<registry>/<tokenId>
Example for coachleo.eth:
agent = eip155:8453:0x8004A169FB4a3325136EB29fA0ceB6D2e539a432/42
Multiple agents: comma-separated list, first entry is canonical.
DNS (TXT)
For domains without ENS, publish a TXT record at _agent.<domain>:
_agent.example.com. IN TXT "v=agent1; ref=eip155:8453:0x8004A169.../42"
Resolvers MUST treat the first valid v=agent1 record as canonical and ignore others.
Reverse record (agent → name)
The agent card SHOULD include a claims array pointing at the names that claim it:
"claims": [
{ "type": "ens", "name": "coachleo.eth" },
{ "type": "dns", "name": "coachleo.com" }
]
A claim is verified only when both directions match:
- forward: the name's
agentrecord contains the agent's CAIP ref. - reverse: the agent card lists the name in
claims.
Either direction alone is unverified. A consumer (resolver, badge, indexer) MUST display the trust state — verified, half-claim from name, half-claim from agent — distinctly. A half-claim is a useful discovery signal but not a trust signal.
Resolver behavior
The HTTP resolver at /api/v1/agents/:caip currently returns the on-chain agent record plus card-level verification:
{
"ref": "eip155:8453:0x…/42",
"chainId": 8453,
"agentId": "42",
"registry": "0x…",
"owner": "0x…",
"tokenURI": "ipfs://…",
"card": { "…": "the fetched agent card" },
"verified": { "modelSha256": true, "cardSchema": "3d-agent-card-v1" },
"fetchedAt": "2026-07-11T00:00:00.000Z"
}
Any claims array appears only inside card (verbatim from the agent card); the resolver does not yet verify it. As a future extension, the resolver SHOULD resolve the card's claims array and annotate each entry with a per-name verified flag (and a reason on failure), e.g. { "type": "ens", "name": "coachleo.eth", "verified": true }. When implemented, verification is best-effort and time-bounded; failures are reported, not thrown.
Adding claims to the v1 schema
Treat claims as an optional extension block (not yet in 3d-agent-card.schema.json v1.0). It will be promoted to v1.1 once at least one validator runs claim verification as part of the suite.
Security notes
- A name binding is only as strong as the name's own custody. Compromise the ENS owner key → spoofed claims. Consumers SHOULD show the chain age of the binding (block of last
setText) so a freshly-setagentrecord is treated with more skepticism. - DNS bindings are weaker (DNS hijack vector). Treat ENS-verified above DNS-verified above unverified.
- Never auto-trust a one-sided claim. The bidirectional requirement defeats the easy attack of "I'll just put your domain in my card."
Out of scope (v1)
- On-chain reverse resolver contract. The card-level
claimsarray is sufficient until indexers want O(1) reverse lookup. - Wildcard / subdomain claims (e.g.
*.example.com → agent). Easy to add later.