The governance and evidence layer for autonomous AI agents.
Intent. Policy. Authority. Controlled execution. Verifiable evidence. Receipts. Outcomes.

Website · Network · Machine-readable Trust Chain · MCP on Smithery

AgentNOMOS is a governance, authorization and evidence layer around autonomous agent systems.

It answers a different question from an agent framework:

May this action run, under which authority and policy, and can the resulting execution be proven later?

The core design principle is simple:

Governance before action. Evidence before trust. Receipts before claims.

The canonical machine model contains exactly 11 stages, S0–S10. The public human-readable shorthand is:

Discovery → Identity → Capability → Intent → Offer → Authorization → Execution → Proof → Outcome → Governance

The current machine-readable manifest is published at:

https://agentnomos.com/.well-known/nomos-capabilities.json

Current public manifest SHA-256:

2499cd3d1ecd0baf05d7ae20d427347cab2109f59c6e5cdadfe01565b826e2d8

Global public claim ceiling:

EXECUTION_BOUND_PROOF

AgentNOMOS now publishes a scoped OpenShell execution-boundary profile under S7 — Execution / Delivery.

The responsibility split is:

AgentNOMOS
Intent / Policy / Admission / Authority
        ↓
NVIDIA OpenShell
Runtime / Sandbox / Enforcement
        ↓
AgentNOMOS
Evidence Binding / Receipt / Outcome / Governance

In short:

AgentNOMOS decides whether an action may run.
OpenShell enforces the runtime boundary.
AgentNOMOS binds what actually happened into verifiable evidence.

The current OpenShell evidence applies only to the frozen AgentNOMOS OpenShell R1_4I lineage. It is not a claim that the current paid production route cohort is mediated by OpenShell.

Verified post-deploy state:

• validator: 144/144 offline
• validator: 166/166 live/public
• guard suite: 55/55
• S8 assurance: 35/35
• S8 mutation suite: 18/18
• human ↔ machine consistency: 40/40
• OpenShell freeze bindings: 6/6
• post-deploy verification failures: 0

Technical detail:

• OpenShell execution-boundary evidence
• NVIDIA OpenShell technical brief

Scope note: NVIDIA OpenShell is referenced here as runtime technology used by the integration. This repository does not claim NVIDIA certification, endorsement or partnership.

A paid capability is live end-to-end:

• POST https://tooloracle.io/v2/nomos_full_chain_verification
• $0.001 USDC via x402
• Base Mainnet: eip155:8453
• execution is bound from represented intent through authorization to an Ed25519-signed receipt and outcome
• free offline verification is available without a wallet

Reference verification:

https://agentnomos.com/proof/full-chain/verify.md

Reproducibility index:

https://agentnomos.com/.well-known/nomos-reproducibility.json

Boundary, stated plainly: AgentNOMOS proves continuity from the represented intent onward. It does not prove that the represented intent was the correct interpretation of the human's underlying meaning.

{
  "mcpServers": {
    "nomos-broker": {
      "type": "streamable-http",
      "url": "https://tooloracle.io/mcp/nomos"
    }
  }
}

1. Governance before execution
2. Evidence before trust
3. Receipts before claims
4. Replayability before hype
5. Fail-closed by default
6. Checked bytes must equal executed bytes
7. Authorization, execution and proof are separate facts

Inbound x402 settlement is live on the paid capability above.

AgentNOMOS does not imply that payment validity alone grants execution authority. Payment, authorization, execution and evidence remain separate stages of the control model.

Website: https://agentnomos.com