AgentNOMOS

Agent trust & execution-proof infrastructure

Govern AI agents before they act.
Prove what happened after.

AgentNOMOS binds intent, capability, authorization and execution into a cryptographically verifiable receipt — and ties the observed outcome back to that chain.

See how it works Verify the proof

A2A · MCP · x402 · Signed receipts · Base Mainnet

§ 1The problem

Capability is easy. Trust is a decision.

¬.01Execution is not proof. An agent can call an API and report success. Without a signed, verifiable record binding the action to its authorization, "it ran" is a claim — not evidence.
¬.02Payment is not authorization. x402 settles who paid. It does not decide whether this agent, under this intent and policy, was allowed to take this action at all.
¬.03Identity is not authority. Knowing who an agent is says nothing about what it may do right now. Authority is a decision — made against current policy and evidence, before execution.

AgentNOMOS is the decision layer in between — and the receipt afterwards.

§ 2How it works

One chain from intent to outcome.

Every governed action passes through the same bound sequence. Each stage is recorded; the signed receipt seals the chain, and the outcome is tied back to it.

S.01IntentWhat the agent is trying to achieve, declared and digested.
S.02CapabilityThe specific advertised capability being invoked.
S.03OfferThe priced offer for this invocation.
S.04Security EvidenceCurrent, source-bound evidence attached to the decision.
S.05AuthorizationThe admission decision for this exact action.
S.06ExecutionThe real action, bound to what was authorized.
S.07Signed ReceiptEd25519 — a signed record of what actually ran.
S.08OutcomeThe observed result, bound back to the chain.

Full chain, stage by stage

§ 3What it does

Four operations. One system.

OP.01 VerifyIs the requested capability real, and is the attached evidence current and validly signed? capability + evidence
→ VALID | INVALID
OP.02 AuthorizeIs this exact action permitted for this agent under the current intent and policy? intent + policy + action
→ PERMIT | DENY
OP.03 ExecuteBind the real execution to the authorized action — nothing more, nothing else. authorized action
→ bound execution
OP.04 ProveReturn a signed receipt the caller can verify offline — public, stateless verifier, published JWKS. No account, no API access, no wallet. Includes Trust Passport — a verifiable status badge for agents and services. execution + outcome
→ SIGNED RECEIPT
§ 4Public proof

Don't trust the diagram. Verify the execution.

A real, paid, governed execution — publicly reproducible.

Capabilitynomos_full_chain_verification
NetworkBase Mainnet · eip155:8453
Price$0.001 USDC · 1000 atomic
Invocationx402
Verifystateless / offline — 21/21 PASS
ReferencePublic ALLOW case · Production DENY case
Receiptsigned · Ed25519
Replay — copy-paste, documented
# Free offline verification — published reference bundles
curl -sO https://agentnomos.com/proof/full-chain/allow-reference.json
curl -sO https://agentnomos.com/proof/full-chain/deny-reference.json
curl -sO https://agentnomos.com/proof/full-chain/nomos_repro_verify.py
curl -s  https://feedoracle.io/.well-known/nomos-execution-jwks.json -o jwks.json

python3 nomos_repro_verify.py --allow allow-reference.json \
                              --deny  deny-reference.json  \
                              --jwks  jwks.json
→ 21 checks: 21 PASS, 0 FAIL, 0 NOT_RUN
Reproduce the proof Open the verifier page

What this proves: the bound execution chain and its reproducible references. What it does not claim: third-party audit of our internal policy.

§ 5Built for

Built for consequential agent actions.

Contexts and the NOMOS role in each
ContextNOMOS role
Agent PaymentsBind authorization to execution — designed to keep payment and execution from drifting apart; each settled call through the chain carries a receipt bound to its authorization.
Autonomous ProcurementPreserve mandate through purchase — bind mandate, purchase, execution and outcome, so a buying agent can show it was allowed to.
Agent MarketplacesVerify authority before invocation — check capability and authorization before invoking another machine, and hold a receipt afterwards.
Regulated WorkflowsAttach evidence to consequential action — current evidence and governance before the action runs, not after.
§ 6Developers

Machine-discoverable by design.

Agents can discover AgentNOMOS without a human in the loop. Humans start here:

InterfaceLocation
Docsagentnomos.xyz/docs ↗
Agent Card (canonical)agentnomos.network/.well-known/agent-card.json ↗
Capabilities/.well-known/nomos-capabilities.json
MCP/.well-known/mcp.json
x402/.well-known/x402.json
AI Catalog/.well-known/ai-catalog.json
llms.txt/llms.txt
Agent Card──▶agentnomos.network ↗Canonical machine identity.
Capabilities──▶.well-known ↗Capability manifest.
MCP──▶.well-known ↗MCP metadata.
x402──▶.well-known ↗Payment manifest.
AI Catalog──▶.well-known ↗Catalog entry.
llms.txt──▶agentnomos.com ↗LLM-readable summary.
§ 7Underneath

One trust chain. Specialized infrastructure underneath.

FeedOracle──▶EvidenceData, regulatory and evidence sources the chain draws on.
ToolOracle──▶Discovery / InvocationCapability access and x402 service routing.
OracleNet──▶Presence / FreshnessMesh and signal transport binding evidence to decisions.
AgentNOMOS──▶Governance / ProofAdmission, authorization and the signed receipt.

These are components of one system — not three products competing for your attention.

§ 8About

AgentNOMOS is trust infrastructure for autonomous machine actions. It connects intent, capability, evidence, authorization, execution and proof so that consequential agent actions remain attributable and verifiable.

Built by FeedOracle Technologies in Germany.

Agents will act autonomously.
Their actions shouldn't become unverifiable.

Explore AgentNOMOS Read the docs Ask NOMOS