AI 4UAnalyze my business
Back to Case Studies
AI + FintechUser-Controlled

FirstAgent

Conversational Payment Authorization

A conversational payment concept that turns a user's request into explicit constraints, a reviewable cart, and a signed authorization step before a transaction can proceed.

Intent
Mandate
Cart
Review
Biometric
Approval
Audit
Trail

Proof boundary: This record documents a product and security architecture. Live transaction volume, transfer coverage, protocol precedence, delivery time, and production readiness have not been independently audited.

Secure Enclave
Cryptographic Auth

Control Before Execution

The product design separates intent, cart review, authorization, and execution so each state can be inspected or revoked.

User-Controlled Flow

How Agent Payments Work

Users describe what they want in natural language. AI extracts rules. Cryptographic mandates authorize agent actions.

1

Describe Mission

User: "Find me wireless headphones under my budget this week"

2

AI Extracts Rules

The rule parser extracts product, maximum price, timing, and quantity

3

Sign Mandate

Secure Enclave + Face ID creates cryptographic auth

4

Action Proceeds

The workflow proposes a cart and proceeds only within the selected approval mode

Security Architecture

Cryptographic Mandates

The architecture binds each proposed action to a cryptographically signed mandate. Two mandate types are represented:

  • Intent Mandate: Authorizes agent to search and propose carts within rules
  • Cart Mandate: Authorizes specific purchase execution (auto or manual)

Secure Enclave Signing

The signing design keeps private-key operations in device hardware and requests biometric authentication before creating a signature.

SigningService.sign(data: Data, with: SecKey) → ECDSA P-256 Signature

Mandate Data Structure

struct IntentMandate {
  let id: UUID
  let missionId: UUID
  let rules: MissionRules        // maxPrice, dates, merchants
  let mode: ExecutionMode        // .ask or .auto
  let signature: Data            // ECDSA P-256
  let publicKey: Data            // Verification key
  let createdAt: Date
  let expiresAt: Date
  var isRevoked: Bool
}

Execution Modes

Ask Mode

Agent proposes carts for user approval. Each purchase requires explicit consent via Face ID. Safer for high-value or unfamiliar purchases.

User approves every transaction

Auto Mode

This concept explores execution within signed mandate rules. Any live implementation would also require merchant permission, risk controls, revocation, monitoring, and a clear recovery path.

Autonomous within constraints

Technical Stack

iOS App
  • Swift 5.9
  • SwiftUI
  • Secure Enclave
  • LocalAuthentication
Intelligence Layer
  • Structured Output
  • Schema Validation
  • Rule Extraction
  • Mission Summary
Payments
  • Payment Orchestration
  • Transfer Contract
  • Transaction Review
  • Multi-currency Data
Security
  • ECDSA P-256
  • Face ID / Touch ID
  • Mandate System
  • Audit Trail

Documented Product Facts

Explicit Intent

A mission record keeps the product, budget, timing, quantity, and approval mode visible before any purchase action.

Hardware-Backed Signing

The architecture pairs device-backed signing with revocable mandates and an audit trail. It does not claim a completed security certification.

Reviewable Transactions

Intent and cart mandates separate what the agent may explore from the specific transaction a user approves.

Building the future of AI payments?

We design agent workflows, payment integrations, and transaction systems around explicit authorization and inspectable state.

Start Your ProjectView More Case Studies