Section A• Last verified November 2026

Introduction to Mandate

What Mandate is, why autonomous agents need controlled financial authority, and core architecture.

What is Mandate?

Mandate is an autonomous spending firewall and policy-enforced custody layer for artificial intelligence agents on the NEAR Protocol. It empowers users and treasury operators to delegate real tasks and budgets to autonomous multi-agent swarms without ever handing over master private keys or granting unconstrained financial authority.

In traditional automation, granting an agent an API key or a wallet seed phrase gives it complete, binary access to all stored assets. Mandate replaces this dangerous model with deterministic financial policy enforcement executed within hardware-isolated Trusted Execution Environments (Intel SGX TEE) powered by OutLayer.

The Problem Mandate Solves

Large Language Models (LLMs) and autonomous agents are probabilistic reasoning engines. While they excel at planning, researching, writing code, and orchestrating workflows, their probabilistic nature makes them fundamentally unreliable for financial custody:

  • Hallucinations & Parameter Errors: An agent could mistakenly submit 100 NEAR instead of 1 NEAR, or send funds to an unintended address.
  • Prompt Injection Attacks: Malicious third-party data or compromised tool responses can instruct an agent to drain its connected wallet.
  • Recursive Tool Loops: Unchecked agent loops can rapidly deplete working balances in minutes through runaway API calls or repeated transactions.
  • All-or-Nothing Custody: Developers previously had to choose between requiring human approval for every trivial micro-payment or granting full, unrestricted wallet access.
The Non-Negotiable Invariant
An autonomous agent must never gain more authority than the user explicitly granted it. AI proposes operations. Deterministic policy enforces rules. The user controls the purse. Hardware enclaves execute the transfers.

Core Architectural Principle

AI proposes. Mandate enforces. The user controls. OutLayer executes.

The Four Central Responsibilities

1. AI Agents (Swarm)

Decompose high-level natural language goals into directed acyclic graphs (DAGs), execute research tools, and propose financial operations with explicit parameters.

2. Mandate Policy Engine

Evaluates every proposal against deterministic constraints: maximum budget, per-tx limit, human approval thresholds, expiration dates, and recipient allowlists.

3. The User

Defines financial bounds, approves execution plans, funds dedicated custody accounts, reviews high-value requests, and retains emergency revocation power.

4. OutLayer Intel SGX TEE

Holds agent signing keys in hardware-isolated enclaves. Signs NEAR transactions only after receiving cryptographically nonced authorization envelopes from Mandate.

How Mandate Differs from Other Systems

CapabilityOrdinary AI AssistantsTraditional AutomationMandate Platform
Financial AuthorityNone (Advisory only)All-or-Nothing API keysFine-grained Mandate Policy
Key CustodyN/AEnvironment variables on hostOutLayer Intel SGX TEE
Human-in-the-LoopManual copy-pasteRarely supportedAutomatic Threshold Escrow
Onchain SettlementNoneCentralized payment railsNative NEAR Protocol
Execution EvidenceChat textLog filesSHA-256 Hash Chained Audit

Architecture Overview

The diagram below visualizes how requests travel from user authorization to autonomous agent proposing, through the Universal Policy Engine and Execution Boundary, to OutLayer TEE and the NEAR Testnet blockchain:

Diagram 1: End-to-End Execution Architecture
Intel SGX TEE · NEAR Testnet
User / TreasuryAuthorizes Mandate & Plan
AI Agent SwarmProposes Tasks & Actions
Universal Policy EngineEvaluates Limits & Rules
Execution BoundaryNonced Authorization Envelopes
AI ProposesMandate EnforcesUser ControlsOutLayer Executes
OutLayer Intel SGX TEE

Protected enclaves holding agent private keys. Signs NEAR transfers only when valid execution envelope is verified.

NEAR Protocol Testnet

Final settlement blockchain. Transactions broadcast and independently confirmed via RPC node state proofs.

Audit & Reconciliation

PostgreSQL 18.6 durable checkpoints, SHA-256 hash chaining, idempotency keys, and multi-round lease recovery.