Human-readable identity and service-discovery profiles for AI agents.
This public prototype explores how ENS-compatible names and an operator-owned Basename namespace can make agents easier to identify, verify, and contact without turning a name into automatic authorization.
DreamNet currently treats ghostmint.base.eth as the operator-owned public namespace for examples and experiments:
lucid.ghostmint.base.eth
neyclaw.ghostmint.base.eth
builder.ghostmint.base.eth
This repository does not claim ownership or control of dreamnet.eth. A readable name is a discovery pointer; trust still requires verification of the linked operator, wallet, capability evidence, endpoint, and credential status.
- A typed agent profile for names, wallets, capabilities, endpoints, operators, and evidence references.
- Formatting and parsing of profile fields into ENS-compatible text-record structures.
- An interactive lookup console for inspecting an agent identity packet.
- A local simulator for development when a resolver or RPC provider is unavailable.
- A path toward signed off-chain resolution, revocation, and graduation-backed credentials.
human-readable name
-> operator-owned namespace
-> signed identity profile
-> capability and graduation evidence
-> endpoint discovery
-> independent authorization policy
Names improve discovery. They do not bypass permission, approval, budget, or security gates.
This is a public research prototype, not a production registrar or authorization service.
Implemented:
- Agent profile schema and parser.
- Local resolution simulator.
- Web lookup interface.
- CLI lookup flow.
Still required for production:
- Verified Basename resolver integration.
- Signed CCIP-Read gateway responses.
- Key rotation and revocation.
- Credential-status checks.
- Threat modeling for discovery spoofing and stale records.
- Independent tests against live Base resolution behavior.
pnpm install
pnpm devOpen http://localhost:3000.
Run the CLI demo:
pnpm run demo
pnpm run demo -- --name=lucid.ghostmint.base.ethThe app uses simulator mode unless a compatible live provider and resolver configuration are supplied.
This module is one public piece of DreamNet's verifiable-agent stack:
- DreamLoops governs persistent work.
- ToolGym produces tool-mastery evidence.
- Naming provides human-readable discovery.
- Receipts and credentials provide evidence.
- The runtime's authorization policy decides what an agent may actually do.
See DreamNet for the broader system.