AGENT TRUST PROTOCOLDeveloper preview Arc-compatible
01 / PERMISSION BEFORE EXECUTION

Autonomy.
Within your limits.

Agents propose the action.
You define what happens next.

A permission layer for agent identity, spending policies and accountable decisions. Explore the flow before connecting a wallet.

Follow the permission path
HUMAN INTENTMACHINE-READABLE BOUNDARIES
THE PERMISSION PATH

Follow one decision.

INTERACTIVE DEMO · NOT LIVE
INPUT / UNSIGNED INTENT

An agent proposes.

ATP requests a 420 USDC vendor payment. A request is not permission to execute.

REQUEST / DEMO-001EVALUATING
Agent
ATP
Intent
420 USDC
Policy limit
2,500 USDC
Amount check
Waiting
Audit entry
Awaiting decision

No wallet signature. No real transaction.

02 / BUILT AROUND ACCOUNTABILITY

Delegate the work.
Not the boundaries.

Three parts of a permission system, designed to stay connected.

01

Know the actor

An agent identity links a request to an owner and a scoped role. Identity is context, not a guarantee of trust.

02

Define the scope

Express limits on amounts, recipients and sessions. Every proposed action can be evaluated against the same rules.

03

Keep the reason

Connect the request to the policy and decision. A useful audit trail explains why an action was allowed or blocked.

03 / DEVELOPER OVERVIEW

A checkpoint.
Not a blank check.

Place policy evaluation between an agent’s intent and the execution boundary. Wallet access is available in the workspace; transaction writes remain gated until contracts are configured.

Open demo workspace ↗
CONCEPTUAL FLOWNOT AN SDK CALL
intent = agent.propose()
policy = owner.boundaries

decision = evaluate(intent, policy)

if decision.allowed:
  // execution boundary
else:
  // stop before signing

audit.record(intent, decision)

Illustrative pseudocode · no execution here

PREVIEW NOTES

Know what you’re exploring.

Interactive, not transactional.The walkthrough and workspace use example identities, balances and decisions. They are not onchain activity.

Integration boundaries.Connecting a wallet does not deploy a policy or authorize agent spending. Contract writes remain gated.

Trust is earned.This preview makes no claim of a deployed, audited protocol or guaranteed protection of funds.