Section C• Last verified November 2026

How Autonomous Spending Works

End-to-end authorization pipeline from proposal to OutLayer TEE signing and NEAR settlement.

Autonomous Financial Execution

Autonomous spending in Mandate is neither an unconstrained agent free-for-all nor a tedious manual workflow where every trivial micro-transfer requires human approval.

Instead, transactions execute autonomously and instantaneously as long as they strictly satisfy the deterministic rules, budgets, and allowlists pre-authorized by the user. If a transaction attempts to exceed limits or transfer to an unknown address, it is either safely blocked or escalated for explicit human approval.

The End-to-End Spending Pipeline

Diagram 2: Autonomous Spending Pipeline
Strict Fail-Closed
STEP 1
Active Mandate
User authorizes budget & limits
STEP 2
Agent Proposal
Specifies recipient, amount & reason
STEP 3
Policy Evaluation
Checks budget, per-tx limit & recipient
ALLOW / ALLOW_WITH_LIMITS

Executes autonomously without human delay

REQUIRE_APPROVAL

Pauses in escrow until human signs approval

DENY / ESCALATE

Operation rejected; budget untouched

STEP 4
Budget Reservation
Funds locked to prevent race conditions
STEP 5
OutLayer TEE Signing
Cryptographic envelope verified
STEP 6
NEAR RPC Settlement
Tx verified & accounting reconciled

Role of ExecutionBoundaryService

The ExecutionBoundaryService is the central choke point of all financial execution in Mandate. It is mathematically impossible for an agent or worker to reach OutLayer TEE or broadcast a NEAR transfer without an execution authorization envelope signed by this service.

  • Atomic Nonce Tracking: Increments a monotonic integer counter per tenant to prevent replay attacks.
  • Idempotency Keys: Guarantees that network retries or worker restarts never trigger duplicate transfers.
  • Tenant Isolation: Validates session credentials to ensure agents can only draw funds from their owner's active mandates.