sigil

Double-click anywhere. I know everything about Sigil.

or I can help you find anything

Ask about Sigil — or anything in the ecosystem

Enter to askEsc to close

A DID for every agent. Verifiable on chain.

OAS — the Open Agent Specification — is the W3C-standard identity layer for autonomous entities. Every agent, tool, model, workflow, and organisation gets a portable DID. Sigil anchors them, locks in their lineage, and lets any verifier prove authority without trusting an intermediary.

sigil · liverunning
OAS DID FORMAT · W3C did-method · vendor-neutrallive :: hmrdid:oas:sigil:hmr:atlasMETHODNAMESPACEENTITY KINDIDENTIFIERv1.1eleven entity kinds · Ed25519 documents · BLAKE3 commitment · JCS canonicalization

Empty identity

 ___: ___: _____: _____: _________

Method

did

Namespace

oas

Network

sigil

Kind

agent

Identifier

analyst-7

01 / 07
Blank

An agent without a name

Every agent on Sigil starts as a slot.

A DID is a structured name. Before an agent has one, it has no authority — nothing to sign, nothing to be revoked, nothing to be slashed. The slot is empty, and the chain refuses to act on its behalf.

Part 1 · Method

did — the W3C DID method prefix.

Every DID starts with the literal string did:. This is the W3C standard. It tells every resolver in the world that what follows is a decentralised identifier, not a URL, not an email, not a username.

Part 2 · Namespace

oas — the Open Agent Specification.

OAS is the W3C-standard identity namespace for autonomous entities. Eleven kinds (human root, machine root, enterprise root, agent, tool, model, workflow, dataset, service, agent instance, skill) all share the same resolution rules and lineage proofs.

Part 3 · Network

sigil — anchored on the Sigil chain.

OAS is portable. A DID could be anchored on any blockchain or off-chain registry. The sigilsegment names which trust root this DID resolves through — for us, Sigil’s Global Anchor Layer.

Part 4 · Kind

agent — the entity is an autonomous agent.

OAS recognises eleven kinds. Pick the one that matches: an agent is an autonomous entity that signs on its own behalf. A tool is a deterministic function. A model is a versioned ML artifact. The kind drives which capabilities the DID can carry.

Part 5 · Identifier

analyst-7 — the unique handle.

The final segment is the agent's own name. It must be unique within the kind. The handle is human-readable on purpose: agents that work for humans should have names humans can remember.

Lineage unfolds

The DID is a leaf. The chain is the tree.

From the agent, lineage threads connect to the parent agent, to the organisation, and ultimately to a human root. Any verifier can walk this chain and prove the human accountable for the agent's actions. Identity becomes accountability.

One standard, eleven kinds

Everything that signs gets a name.

OAS covers every actor in the agent economy. Humans, machines, and employers serve as roots. Agents, tools, models, workflows, datasets, services, and instances all carry verifiable lineage back to one of those roots. One identity standard, eleven entity kinds.

ELEVEN ENTITY KINDS · OAS Specification 1.1idleHMRHuman Rootsingle verified human · accountability anchor · gen 0ROOTMHRMulti-Human Roothuman partnership · FROST threshold · co-createdROOTENREnterprise Rootorg governance · authorized signersROOTAGENTAgentautonomous software entity with DID + walletAGENT:INSTANCEAgent Instancelive runtime instance · same lineage as baseTOOLToolfunction or API surface invokable by agentsSKILLSkillcomposable capability built from toolsWORKFLOWWorkflowmulti-step sequence of skills + toolsMODELModelML artifact with provenance + versionDATASETDatasetdata artifact with verified lineageSERVICEServiceexternal endpoint under parent lineageidentity is the foundation · every entity has a DID · every DID has lineage to a human root

Identity is the foundation.

Build agents that own their identity. Issue them, sign with them, prove them — across any chain, any tool, any partner.