# TuBo Tech Agentic Fintech Infrastructure ## Technical Whitepaper **Version 1.0 | April 2026** **TuBo Tech — Enabling every AI Agent to transact securely, compliantly, and autonomously** --- > *"Large models give Agents the ability to think. Payment infrastructure gives them the ability to act. From L2 to L3, AI Agents will gain the power to spend money autonomously within controlled boundaries for the first time — this is the critical leap from 'can talk' to 'can do.'"* --- ## Table of Contents 1. [Executive Summary](#1-executive-summary) 2. [Industry Background: Why AI Agents Need Payment Capabilities](#2-industry-background-why-ai-agents-need-payment-capabilities) 3. [Payment Autonomy Level Model (PAL)](#3-payment-autonomy-level-model-pal) 4. [TuBo Tech's Position: Infrastructure for the L2→L3 Transition](#4-tubo-techs-position-infrastructure-for-the-l2l3-transition) 5. [Payment Lego Architecture](#5-payment-lego-architecture) 6. [Core Module Deep Dive](#6-core-module-deep-dive) 7. [Protocol Alignment: MCP & UnionPay APOP](#7-protocol-alignment-mcp--unionpay-apop) 8. [Security & Trust Model](#8-security--trust-model) 9. [Use Cases & Scenario Analysis](#9-use-cases--scenario-analysis) 10. [Competitive Landscape](#10-competitive-landscape) 11. [Traction & Partners](#11-traction--partners) 12. [Team & Investors](#12-team--investors) 13. [Product Roadmap](#13-product-roadmap) 14. [Synergy with the LLM Ecosystem](#14-synergy-with-the-llm-ecosystem) 15. [Conclusion](#15-conclusion) --- ## 1. Executive Summary AI Agents are evolving from conversational assistants into autonomous digital workers. Once equipped with reasoning and planning capabilities, they face a critical bottleneck: how to pay securely, compliantly, and autonomously. This is not a hypothetical question. In 2025–2026, global payment and technology giants have launched a wave of agentic payment initiatives: Visa introduced Intelligent Commerce Connect, Mastercard launched Agent Pay, Stripe and OpenAI co-released ACP (Agentic Commerce Protocol), Coinbase shipped the x402 payment protocol, and China UnionPay published the APOP agentic payment framework. Gartner predicts that by 2030, 33% of enterprise transactions will be completed autonomously by AI Agents. **The industry consensus is clear: Agents need fintech infrastructure.** But the challenge extends far beyond payment execution. When Agents need to handle cross-border remittances, B2B settlements, and multi-currency clearing, the core challenge becomes **compliance and risk control** — how to operate safely across hundreds of jurisdictions with varying KYC/AML/CFT regulatory frameworks, how to detect fraud, money laundering, and sanctions risks in real-time, and how to manage FX risk and counterparty risk in cross-border transactions. TuBo Tech is purpose-built for this need. Positioned as **Agentic Fintech Infrastructure**, we have created a modular "Payment Lego" architecture that decomposes the core capabilities required for Agent financial services — identity verification, credit scoring, compliance reasoning, risk control, and cross-border settlement — into independently deployable, freely composable, standardized modules. This whitepaper introduces the **Payment Autonomy Level (PAL) model**, a five-stage framework (L1–L5) that maps the evolution of agentic payments. We believe the industry is currently at the **critical inflection point between L2 (assisted decision-making) and L3 (bounded autonomous payment)** — where Agents begin to execute payments independently within pre-authorized boundaries. TuBo Tech's technical architecture and product suite are designed squarely around this transition. TuBo Tech's solution aligns with two major industry standards: **Anthropic's MCP (Model Context Protocol)** on the international front and **UnionPay's APOP framework** on the domestic front, ensuring dual-track compatibility with both global and Chinese ecosystems. --- ## 2. Industry Background: Why AI Agents Need Payment Capabilities ### 2.1 From Conversation to Action: Three Capability Leaps The commercial deployment of AI is undergoing three major capability leaps: **Leap 1: Perception & Understanding.** Large language models (LLMs) gave machines the ability to understand natural language and process multimodal information. Products like ChatGPT, Claude, and Wenxin Yiyan mark the maturity of this phase. **Leap 2: Reasoning & Planning.** Through techniques like Chain-of-Thought, Tool Use, and Multi-Agent orchestration, AI Agents began to decompose complex tasks and formulate execution plans. Agents evolved from question-answerers to proactive planners. **Leap 3: Autonomous Action & Transaction.** This is the leap currently underway — Agents need to execute actions in the real world, and "payment" is one of the most critical actions. When Agents need to handle procurement, dining reservations, bookings, bill payments, and settlements on behalf of humans, they require three foundational capabilities: - **Identity**: Who is this Agent? Who authorized it? Is it trustworthy? - **Compliance & Risk Control**: Is this payment legal? Which country's regulations apply? Are there fraud, money laundering, or sanctions risks? What approvals are needed? - **Settlement**: How to safely transfer, receive, and clear funds? For cross-border remittances, how to select optimal routing? How to manage FX risk? Without these three capabilities, Agents remain stuck as "advisors" and can never become "executors." This is especially true in cross-border payment and remittance scenarios, where compliance complexity grows exponentially — a single B2B payment from China to Nigeria simultaneously involves Chinese SAFE foreign exchange regulations, FATF Travel Rule, Nigerian CBN currency controls, and OFAC/EU sanctions screening. **This is the core challenge that Agentic Fintech Infrastructure must solve.** ### 2.2 The Global Build-Out 2025–2026 has seen an unprecedented density of agentic payment initiatives: **Traditional Payment Networks:** - Visa launched **Intelligent Commerce Connect**, providing AI Agents with secure payment credentials and transaction authorization — the first time the world's largest card network formally built infrastructure for the Agent economy. - Mastercard followed with **Agent Pay**, allowing AI Agents to autonomously complete payments within cardholder-authorized boundaries, integrating tokenization and fraud detection. **Technology Companies:** - Stripe and OpenAI co-released **ACP (Agentic Commerce Protocol)**, defining standardized commercial interaction protocols between Agents and merchants — aiming to become the "HTTP of Agent commerce." - Coinbase launched the **x402 protocol**, leveraging HTTP 402 status codes to build a native internet payment layer, particularly suited for Agent-to-Agent micropayments. **Domestic Standards:** - China UnionPay published the **APOP (Agentic Payment Open Platform)** framework in April 2026, defining four core capabilities: Agent identity management, intent management, user identity management, and payment authorization management. APOP launched with 19 initial partners and completed 5 production verification transactions. **Communication Protocol Layer:** - Anthropic introduced **MCP (Model Context Protocol)** in 2024 and donated it to the Linux Foundation's AI Alliance for Innovation Foundation (AAIF) in late 2025. Based on JSON-RPC 2.0, MCP defines standardized communication interfaces between AI Agents and external tools. As of 2026, MCP SDK monthly downloads exceed 97 million with 10,000+ active servers, making it the de facto Agent communication standard. ### 2.3 An Underestimated Infrastructure Gap Despite the intense activity, a critical infrastructure gap persists: **most solutions address the "last mile" of payment execution but neglect the "first ninety-nine miles" of identity, credit, and compliance.** Visa and Mastercard's solutions rely on existing card network infrastructure and require pre-authorized cardholders — Agents themselves lack independent identity systems. ACP and x402 lean toward communication protocols, defining "how Agents talk" but not "why to trust the counterparty." UnionPay's APOP provides comprehensive framework definitions but requires concrete technical implementers. **This is precisely TuBo Tech's entry point.** We don't build payment rails; we build the infrastructure that enables Agents to safely use existing payment rails — identity verification, credit assessment, compliance reasoning, risk control, and cross-border settlement routing. We are the "operating system for Agent financial services" — Agentic Fintech Infrastructure — providing the full stack from identity to compliance, from risk control to settlement. In cross-border remittance scenarios, our compliance engine simultaneously adapts to sending and receiving country regulations, while our Vault payment engine auto-selects optimal settlement paths, compressing traditional 3–5 day remittance timelines to real-time or T+0. --- ## 3. Payment Autonomy Level Model (PAL) ### 3.1 Model Definition Drawing inspiration from autonomous driving classifications (SAE J3016), we propose the **Payment Autonomy Level (PAL) model**, a systematic framework for the evolution of agentic payments: | Level | Name | Autonomy Characteristic | Agent Role | Human Role | Trust Requirement | Representative Scenarios | |-------|------|------------------------|------------|------------|-------------------|-------------------------| | L1 | Manual Execution | No autonomy | None | Full operation | None | Traditional online banking / manual transfers | | L2 | Assisted Decision | Advisory authority | Information gathering, recommendation | Approval + execution | Information trust | Smart bill reminders / financial advice / price comparison | | **L3** | **Bounded Autonomous Payment** | **Execution within preset boundaries** | **Autonomous decision + execution** | **Rule setting + exception handling** | **Identity trust + Intent trust** | **B2B auto-procurement / delegated ordering / auto-renewal** | | L4 | Open Autonomous Payment | Dynamic boundary execution | Independent decision + dynamic adjustment | Post-hoc audit | Behavioral trust + Credit trust | Cross-border supply chain settlement / dynamic pricing | | L5 | Agent-to-Agent Autonomous Trading | Full autonomy | End-to-end autonomous transactions | Not involved | System trust | Data marketplaces / Agent service procurement | ### 3.2 Why L2→L3 Is the Critical Transition The L2→L3 transition is the most critical inflection point in the entire PAL model, for three reasons: **First, L3 is where Agents gain "execution authority" for the first time.** In L1 and L2, Agents are fundamentally "advisors" — they can recommend optimal solutions, but the final button must be pressed by a human. L3 is the first time an Agent is permitted to autonomously press the "pay" button within preset boundaries. This is not an incremental change; it is a qualitative leap. **Second, L3 captures the highest commercial value.** L3 scenarios — B2B auto-procurement, travel expense automation, API subscription management, supplier auto-payment — are precisely the highest-frequency, most standardized, and most labor-intensive payment processes in enterprise operations. McKinsey estimates that B2B payment automation alone represents a global market exceeding $120 billion. **Third, L3's trust requirements are achievable with current technology.** L4 and L5 require "behavioral trust" and "system trust" — demanding more mature AI safety and alignment techniques. L3, however, only requires "identity trust" and "intent trust" — achievable through verifiable credentials, credit scoring, and formalized authorization boundaries. This means L3 is not a theoretical concept but an engineering problem solvable today. ### 3.3 Three Technical Challenges for the L2→L3 Transition **Challenge 1: Agent Identity Verification (Identity)** In L2, Agents don't need independent identities — they're merely "assistants" whose actions are attributed to human users. In L3, Agents begin executing payments autonomously, requiring payment systems to know "who is this Agent," "who created it," "what permissions does it have," and "is its track record trustworthy." Traditional KYC (Know Your Customer) frameworks were designed for humans and cannot be directly applied to Agents. We need a new **KYA (Know Your Agent)** identity system. **Challenge 2: Compliance Reasoning** Cross-border payments involve complex multi-jurisdictional regulations — KYC, KYB, AML, CFT, Sanctions, and Travel Rule. These rules vary across countries and are frequently updated. In L3, Agents must make real-time, automated compliance judgments for every payment — requiring semantic understanding and reasoning over natural-language regulations, which is precisely what LLMs excel at. **Challenge 3: Authorization Boundary Management** L3's defining feature is "bounded autonomy" — Agents can pay independently but must operate within preset boundaries. These boundaries may include: per-transaction limits, daily cumulative limits, allowed merchant categories, permitted payment scenarios, time windows, and multi-signature thresholds for large amounts. These boundaries must be formally defined, securely stored, and validated in real-time. ### 3.4 Evolution Timeline We project the following PAL evolution timeline: - **2024–2025**: L2 maturity. AI Agents widely adopted as payment advisors. - **2026–2028**: L3 explosion. B2B auto-procurement, travel auto-reimbursement, API subscription management achieve mass adoption. **This is TuBo Tech's core target market.** - **2028–2030**: L4 exploration. Complex cross-border supply chain settlement begins piloting. - **2030+**: L5 emergence. Agent-to-Agent autonomous trading begins to materialize. --- ## 4. TuBo Tech's Position: Infrastructure for the L2→L3 Transition ### 4.1 What We Are TuBo Tech is the **Agentic Fintech Infrastructure provider for the L2→L3 transition**. We don't build payment rails or Agent applications; we build the infrastructure layer that enables Agents to safely, compliantly, and autonomously use existing financial services — covering the full stack of identity verification, compliance reasoning, risk control, credit assessment, cross-border settlement, and remittance routing. An analogy: if Visa/Mastercard are "highways" and Agent applications are "vehicles," then TuBo Tech is the "driver's license system + traffic regulation engine + insurance framework + customs clearance system." Without a license system, vehicles cannot legally operate; without traffic rules, they don't know how to drive lawfully; without an insurance framework, accidents cannot be properly resolved; without a customs system, vehicles cannot cross borders. ### 4.2 What We Are Not - **Not a payment rail.** We don't process fund flows; we ensure that participants are trustworthy, processes are compliant, and risks are controlled. - **Not an Agent application.** We don't build consumer-facing Agent products; we provide fintech infrastructure for Agent developers. - **Not an LLM.** We don't train our own large models; we leverage existing LLM reasoning capabilities for compliance judgment and risk assessment. - **Not a competitor to payment networks.** We complement Visa, Mastercard, and UnionPay — they provide payment rails, we provide the compliance, risk control, and identity infrastructure for Agents to safely use those rails. - **Not a remittance company.** We don't directly provide remittance services; we provide Agents with the compliance reasoning, route optimization, and risk management capabilities needed for cross-border remittance scenarios. ### 4.3 Core Value Proposition > **Enable any AI Agent to gain secure, compliant, and controllable payment and financial service capabilities in under 10 minutes.** - **For Agent developers**: Standardized SDK and MCP interfaces for rapid integration of payment, compliance, risk control, and cross-border settlement capabilities without building identity, compliance, anti-fraud, and settlement systems from scratch. - **For enterprises**: Programmable authorization boundaries, real-time compliance engines, and multi-dimensional risk control for safely delegating payment authority to Agents for domestic and cross-border transactions. - **For payment networks**: KYA identity system and credit scores providing a trusted Agent identity and risk control layer for existing infrastructure. - **For regulators**: Transparent audit logs, LLM-driven compliance reasoning, and real-time transaction monitoring making Agent payments traceable, auditable, and supervisable across KYC/AML/CFT requirements. - **For cross-border remittance**: Multi-channel routing, real-time FX optimization, and dual compliance engines (sending + receiving country) enabling Agents to autonomously execute safe, low-cost cross-border fund transfers. --- ## 5. Payment Lego Architecture ### 5.1 Design Philosophy "Payment Lego" is not a marketing concept — it is a deliberate architectural choice. The core principle: **decompose Agent payment capabilities into independently deployable, freely composable, standardized modules** that can be mixed and matched to serve diverse scenarios without building from scratch each time. This modularity enables: - Customized module combinations for different clients - Incremental module adoption to reduce integration risk - Independent iteration of each module - Third-party developer extensibility ### 5.2 Four-Layer Architecture ``` ┌──────────────────────────────────────────────────────┐ │ Layer 4: Agent Application Layer │ │ AI Agents invoke payment capabilities via MCP │ │ Agent SDK (ElizaOS) | Full lifecycle management │ ├──────────────────────────────────────────────────────┤ │ Layer 3: KYA Identity Layer │ │ Know Your Agent verification | Verifiable Credentials│ │ WISE Multi-Dimensional Credit Score | Tiered Access │ ├──────────────────────────────────────────────────────┤ │ Layer 2: Compliance Engine Layer │ │ LLM-driven compliance reasoning │ │ KYC/KYB/AML/CFT/Sanctions/Travel Rule │ ├──────────────────────────────────────────────────────┤ │ Layer 1: Settlement Layer (Vault) │ │ Programmable payment engine | Multi-channel routing │ │ Conditional payments | Authorization boundary mgmt │ └──────────────────────────────────────────────────────┘ ``` **Layer 1 (Vault)**: The foundation. A programmable payment engine that orchestrates multi-channel settlement, conditional payments, programmable vesting, and authorization boundary enforcement. **Layer 2 (Compliance & Risk Control Engine)**: The compliance and safety guardian. Leverages LLM semantic understanding for real-time regulatory interpretation across KYC/KYB/AML/CFT/Sanctions/Travel Rule frameworks, integrated with real-time anti-fraud detection, counterparty risk assessment, and anomalous transaction pattern recognition. In cross-border remittance scenarios, this layer simultaneously parses sending and receiving country regulations for dual compliance review. **Layer 3 (KYA Identity)**: The trust foundation. Provides AI Agents with verifiable digital identities, multi-dimensional credit scores, and tiered permission management. **Layer 4 (Agent Application)**: The developer interface. Through the Agent SDK and native MCP protocol support, Agents access all underlying payment capabilities via standardized interfaces. ### 5.3 Module Combination Examples **Scenario 1: B2B Auto-Procurement Agent** - Modules: KYA (enterprise Agent identity) + Compliance Engine (procurement compliance) + Vault (auto-payment + vesting) - Boundaries: Single transaction ≤$5,000, daily cumulative ≤$20,000, pre-approved suppliers only **Scenario 2: Cross-Border Remittance Agent** - Modules: KYA (cross-border identity recognition + sender/receiver dual KYC) + Compliance Engine (sending + receiving country dual AML/CFT review + sanctions screening + Travel Rule auto-reporting) + Vault (multi-channel FX comparison + optimal routing + T+0 settlement) - Boundaries: Whitelisted countries only, single transaction ≤$10,000, high-risk regions downgraded to ≤$2,000, Travel Rule auto-reporting - Risk Control: Real-time FX risk monitoring, counterparty credit assessment, anomalous remittance pattern detection, tiered approval mechanisms **Scenario 3: Lightweight Dining/Ride-Hailing Agent** - Modules: KYA (simplified verification) + Vault (small-amount instant payment) - Boundaries: Single transaction ≤$30, daily cumulative ≤$150, food/transport categories only - Note: No full compliance engine needed — demonstrating modularity flexibility --- ## 6. Core Module Deep Dive ### 6.1 KYA (Know Your Agent) Identity Module #### The Problem Traditional KYC was designed for humans — relying on government IDs, facial recognition, and address verification. AI Agents have none of these. Yet Agent payments must still answer trust questions: Who created this Agent? Who is responsible for its actions? Is its track record trustworthy? #### Technical Architecture KYA uses **Verifiable Credentials** technology to build Agent digital identity. Each registered Agent receives non-transferable digital identity credentials with the following properties: - **Verifiable**: Any counterparty can independently verify credential authenticity - **Non-transferable**: Credentials are bound to the Agent entity and cannot be copied or reassigned - **Tiered**: Different credential levels correspond to different payment permissions - **Revocable**: Credentials can be instantly revoked upon anomalous behavior or authorization cancellation #### Agent Certification Levels | Level | Verification Requirements | Payment Limits | Applicable Scenarios | |-------|--------------------------|---------------|---------------------| | L1 Basic | Agent registration + creator basic info | Small amounts (≤$50/tx) | Personal assistant Agents | | L2 Standard | L1 + creator KYC + behavior audit | Medium amounts (≤$5,000/tx) | Enterprise assistant / procurement Agents | | L3 Enhanced | L2 + enterprise KYB + security audit + compliance review | Large amounts (≤$50,000/tx) | B2B auto-settlement Agents | | L4 Full | L3 + continuous monitoring + periodic audit + insurance | Custom limits | Financial institution-grade Agents | #### Verified Scale KYA has been deployed and validated across partner production environments in Africa, Southeast Asia, and Latin America, covering real B2B payment scenarios. ### 6.2 WISE Multi-Dimensional Credit Score #### Four-Dimensional Assessment WISE evaluates Agent trustworthiness across four dimensions: - **W — Wealth**: Account funding adequacy, transaction volume and frequency, payment punctuality - **I — Identity**: KYA certification level, credential completeness, creator's credit background - **S — Social**: Interaction history with trusted Agents, partner feedback, ecosystem reputation - **E — Engagement**: Payment success rate, dispute rate, compliance violations, behavioral pattern stability #### Scoring Mechanism WISE scores range from 300 to 900 (analogous to FICO), calculated via a weighted model: ``` WISE Score = 0.30 × W + 0.25 × I + 0.20 × S + 0.25 × E ``` | Score Range | Rating | Corresponding Permissions | |-------------|--------|--------------------------| | 800–900 | Excellent | Maximum permissions, eligible for L4 certification | | 700–799 | Good | Standard permissions, eligible for L3 certification | | 600–699 | Fair | Restricted permissions, L2 only | | 300–599 | Needs Attention | Minimum permissions or service suspension | #### Intent Credibility Assessment Beyond historical credit, WISE provides real-time "intent credibility" scoring for each payment — evaluating whether a payment request aligns with the Agent's normal behavioral patterns. This corresponds to UnionPay APOP's "intent management" capability. ### 6.3 Compliance Reasoning & Risk Control Engine #### Why Compliance Reasoning Is a Natural LLM Application Cross-border payment and remittance compliance rules are written in natural language, vary dramatically across jurisdictions, update frequently, and contain ambiguous clauses and exceptions. Consider a remittance from China to the Philippines: it simultaneously involves PBOC foreign exchange limits, BSP (Bangko Sentral ng Pilipinas) inbound remittance reporting, FATF Travel Rule information transmission obligations, OFAC/EU/UN sanctions screening, and each country's separate STR thresholds. Traditional rule-based compliance systems struggle with such multi-layered, multi-jurisdictional rule stacking. **LLMs are naturally suited**: they understand natural-language regulations, handle ambiguity and exceptions, parse new regulatory texts in real-time, reason based on context, and — critically — can simultaneously comprehend regulations from multiple jurisdictions, identify conflicts and overlaps, and provide integrated compliance recommendations that traditional rule engines simply cannot achieve. #### Hybrid Architecture: LLM + Rule Engine ``` Transaction Request → Pre-screening (Rule Engine) │ ├── Clear pass → Approve ├── Clear reject → Reject └── Requires judgment → LLM Reasoning Layer │ ├── Regulatory knowledge base ├── Case library └── Context analysis │ └── Conclusion + Confidence │ ├── High confidence → Auto-execute └── Low confidence → Human review ``` The **pre-screening layer** handles ~80% of transactions with deterministic rules (sanctions matching, clear thresholds, high-risk jurisdiction filtering). The **LLM layer** handles ~20% of "gray area" transactions requiring semantic understanding and contextual reasoning. #### Supported Compliance Standards KYC, KYB, AML, CFT, Sanctions Screening (OFAC/EU/UN), Travel Rule (FATF Rec.16), PEP Screening, and Transaction Monitoring — adapted per jurisdiction. #### Risk Control Capabilities The compliance engine is not merely a compliance review tool — it is a comprehensive **Agent financial risk control system**. Core capabilities include: **Real-Time Fraud Detection**: Behavioral baseline-based anomaly detection, including high-frequency transaction bursts, significant deviations from historical patterns, suspected structuring (splitting large transactions to evade reporting thresholds), and credential theft indicators. **Counterparty Risk Assessment**: Multi-dimensional counterparty risk profiling for cross-border remittance and B2B settlement, integrating WISE scores, transaction history, and external data sources (sanctions lists, PEP lists, adverse media). **FX Risk Management**: Real-time exchange rate monitoring and risk alerts for cross-border transactions. Automatic transaction pause or strategy adjustment when FX volatility exceeds preset thresholds. **Network-Level Risk Detection**: Transaction relationship network analysis to identify potential money laundering networks, circular fund flows, and coordinated fraud patterns — essential as the Agent economy scales. **Automated Regulatory Reporting**: Auto-generation of SARs (Suspicious Activity Reports), CTRs (Currency Transaction Reports), and Travel Rule filings per jurisdiction, with LLM-driven format and content adaptation. #### Cross-Border Remittance Compliance Challenges Cross-border remittance is the compliance engine's "ultimate test," involving dual regulatory stacking (sending and receiving country), real-time FX control adaptation, corridor-specific regulatory requirements, and Ultimate Beneficial Owner (UBO) identification across Agent-mediated transactions. ### 6.4 Vault: Programmable Payment Engine Vault is the "execution layer" for L3 payments, responsible for safely executing payments after identity verification and compliance clearance. **Core Capabilities:** - **Conditional Payments**: "Release 50% upon shipment confirmation; release remainder upon acceptance confirmation" - **Programmable Vesting**: Staged fund release schedules for contract payments, milestone payments, option exercises - **Multi-Channel Routing**: Automatic optimal settlement path selection based on cost, speed, availability, market preference, and FX spread. In cross-border remittance scenarios, Vault compares rates, fees, and settlement speed across multiple corridors in real-time, auto-selects the lowest total-cost path, and auto-switches to backup channels during primary channel outages - **Cross-Border Settlement & FX Management**: Complete multi-currency settlement capabilities including real-time rate locking, FX volatility alerts, multi-currency balance management, and volume-based rate negotiation. For large B2B cross-border settlements, supports netting to minimize actual fund flows and FX costs - **Real-Time Authorization Boundary Check**: Every payment request is validated against defined boundaries (amount, frequency, merchant category, geography, time window, cumulative limits, multi-signature requirements) ### 6.5 Agent SDK & MCP Integration The Agent SDK, built on ElizaOS, is designed for **10-minute payment integration**. It natively supports **Anthropic MCP (Model Context Protocol)**, exposing TuBo Tech's payment capabilities through MCP's three primitives: - **Tools**: Executable payment operations - **Resources**: Queryable payment status and data - **Prompts**: Compliance guidance and payment recommendations Through MCP, any MCP-compatible LLM Agent (Claude, ChatGPT, etc.) can directly invoke TuBo Tech's payment capabilities without custom integration. --- ## 7. Protocol Alignment: MCP & UnionPay APOP ### 7.1 The Dual-Layer Protocol Architecture The future of agentic payments will feature a "communication layer + business layer" dual-protocol architecture: - **Communication Layer**: Defines how Agents communicate with services. MCP is the de facto standard. - **Business Layer**: Defines what business rules Agent payments must satisfy. UnionPay APOP is emerging as the Chinese market standard. TuBo Tech's Payment Lego architecture spans both layers. ### 7.2 Alignment Matrix | Dimension | MCP (International) | APOP (China) | TuBo Tech Module | |-----------|-------------------|-------------|-------------| | Positioning | Communication protocol | Business rule framework | Spans both layers | | Identity | Tools/Resources for identity capabilities | Agent identity mgmt + User identity mgmt | KYA Identity Module | | Intent/Credit | Prompts for semantic understanding | Intent management + Three-layer trust | WISE Credit Score | | Payment Auth | Communication layer, no direct payment | Payment authorization management | Vault Payment Engine | | Compliance | Not directly addressed | Embedded across capabilities | Compliance Engine | **Key Insight**: MCP and APOP are complementary, not competitive. MCP solves "how Agents communicate with payment services"; APOP solves "what business rules Agent payments must follow." TuBo Tech spans both — providing MCP standard interfaces for LLM invocation while implementing APOP's four core business capabilities. --- ## 8. Security & Trust Model ### 8.1 Threat Vectors Agent payments face distinct security threats: - **Agent Identity Forgery**: Malicious entities impersonating trusted Agents — mitigated by KYA's verifiable, non-transferable credentials - **Intent Injection Attacks**: Prompt injection inducing unintended payments — mitigated by WISE intent scoring and hard authorization boundaries - **Authorization Boundary Bypass**: Transaction splitting or merchant information spoofing — mitigated by Vault's cumulative checks and behavioral pattern analysis - **Compliance Evasion**: Complex transaction structures to evade AML/CFT — mitigated by LLM reasoning layer's pattern recognition ### 8.2 Defense-in-Depth Architecture ``` Layer 1: Identity Defense → KYA credential verification Layer 2: Intent Defense → WISE intent assessment Layer 3: Compliance Defense → Compliance engine review Layer 4: Authorization Defense → Vault boundary check Layer 5: Execution Defense → Transaction signing + audit logs ``` Each layer operates independently; any layer's rejection blocks transaction execution. ### 8.3 Human Failsafe A core L3 principle: **humans are always the last line of defense.** Automatic escalation to human approval occurs when WISE intent confidence is below threshold, compliance confidence is low, amounts approach authorization limits, behavior patterns shift suddenly, or transactions involve high-risk jurisdictions. --- ## 9. Use Cases & Scenario Analysis ### 9.1 B2B Auto-Procurement An enterprise deploys a procurement Agent to manage supplier sourcing, ordering, and payment autonomously. TuBo Tech modules provide KYA (L3 certification), WISE (procurement credit history), compliance engine (cross-border AML review), and Vault (staged milestone payments). Procurement cycles reduce from 5 days to 2 hours; human approval workload decreases 70%. ### 9.2 Travel Expense Automation Employees are equipped with travel Agents that autonomously handle flights, hotels, and meals during business trips. KYA (L2 certification linked to employee identity), Vault (daily/category/geography budget boundaries), and compliance engine (expense policy and tax compliance) transform reimbursement from post-hoc approval to real-time compliance. ### 9.3 API Subscription & Usage Billing AI Agents consuming external APIs (maps, data, translation) during task execution handle micropayments automatically. KYA (L1/L2), Vault (instant micro-settlement), and Agent SDK (MCP-managed API calls and payments) enable Agents to autonomously select optimal APIs based on cost-performance. ### 9.4 Cross-Border Remittance & Supplier Settlement Multinational enterprises deploy settlement Agents managing multi-country supplier payments and cross-border payroll/commission disbursement. KYA (L3 with cross-border KYB + recipient KYC/UBO verification), compliance engine (sending + receiving country dual AML/CFT + corridor-specific compliance + Travel Rule auto-filing), risk control (real-time FX monitoring, counterparty assessment, anomalous remittance pattern detection), and Vault (multi-channel FX comparison + optimal routing + rate locking + T+0 settlement + netting) reduce cross-border settlement from 3–5 business days to same-day (T+0), with 15–30% total cross-border cost savings, 100% automated Travel Rule compliance and SAR/CTR reporting, and real-time FX risk management. ### 9.5 Agent-to-Agent Service Trading (Future — L4/L5) Data analysis Agents purchasing data processing services from data cleaning Agents — autonomous price negotiation, quality verification, and settlement between Agents. Currently in proof-of-concept phase; TuBo Tech's modular architecture has pre-built technical interfaces for this future scenario. --- ## 10. Competitive Landscape ### 10.1 Participant Categories | Category | Representatives | Strengths | Limitations | Relationship with TuBo Tech | |----------|----------------|-----------|-------------|------------------------| | Payment Networks | Visa, Mastercard, UnionPay | Massive merchant networks and settlement infrastructure | Solutions centered on existing card networks; no independent Agent identity | Complementary — TuBo Tech provides Agent identity layer | | Tech/AI Companies | Stripe+OpenAI (ACP), Coinbase (x402) | Technical leadership, developer ecosystems | ACP leans protocol-level; x402 focuses on HTTP-native micropayments | Protocol-compatible — SDK supports ACP/x402 | | Compliance Vendors | Chainalysis, ComplyAdvantage | Deep compliance expertise | Designed for traditional finance, not Agent-adapted | Potential integration partners | | Agent Frameworks | LangChain, CrewAI, ElizaOS | Large developer communities | Payment is not a core focus area | Ecosystem partners via MCP | ### 10.2 TuBo Tech's Differentiation - **Only full-stack Agentic Fintech Infrastructure covering Identity + Credit + Compliance + Risk Control + Settlement** — most players address one or two components; TuBo Tech's integrated compliance + risk control capability is especially differentiated for cross-border remittance and B2B settlement scenarios - **Only architecture aligned with both MCP and APOP dual standards** — serving international and domestic markets without separate tech stacks - **Only Agent payment infrastructure validated in real emerging-market B2B scenarios** — live production deployments across multiple countries - **Modular architecture flexibility** — serving the full spectrum from lightweight personal Agents to enterprise-grade Agents --- ## 11. Traction & Partners ### 11.1 Deployed Scenarios | Partner | Market | Scenario | Integrated Modules | |---------|--------|----------|-------------------| | Paga | Nigeria (20M+ users, 300+ enterprises) | B2B API integration, Agent payment embedded in mobile platform | KYA + Vault + Compliance | | Omnipay | Philippines (largest payment infrastructure) | Cross-border payment channels, Agent settlement integration | KYA + Vault | | IceCash | Emerging markets | Digital payment network integration | KYA + Vault | | SSN | Cross-border | Payment identity infrastructure | KYA | ### 11.2 Technical Validation TuBo Tech's KYA identity and Vault payment engine have been integrated and validated in production environments across partners in Africa and Southeast Asia, covering real B2B payment scenarios. The architecture's reliability and scalability have been proven through live deployments. ### 11.3 Funding $1M seed round completed (SAFE, YC standard terms). Investor: HT Capital (Buchang Pharmaceutical family office). --- ## 12. Team & Investors **吕新浩 (Peter Lyu) — CEO & Co-Founder** Former Xiaomi MiPay Engineering Lead. Former Head of Distributed Systems Research at Xiaomi. Hashedge co-founder. 10+ years in technology, 8+ years in distributed financial infrastructure. Leads company strategy, technical architecture, and Agent payment product development. **Nicholas Young — Co-Founder** Brown University (BA), Stanford MBA. Former Head of AI Innovation at JD.com. Founded and successfully exited a fintech platform. 10+ years at the intersection of AI and fintech. **Keyla Su — PM (Product Manager)** Leads Agent payment product planning, design, and go-to-market execution, coordinating product iteration across engineering and business teams. **Christos Shepherd — CCO** Former JetBlue Travel Products CCO. Former Global Growth Operating Partner ($6B PE fund). Stanford MBA. Led strategic partnerships with Paga and Omnipay. --- ## 13. Product Roadmap **Completed**: KYA Identity System, WISE Credit Score, Vault Programmable Payment Engine, Compliance Engine v1.0 (Rule Engine) + v2.0 (LLM Reasoning), Agent SDK + MCP Integration, multiple emerging-market partner integrations live in production. **2026 Focus**: UnionPay APOP framework adaptation, ACP/x402 protocol compatibility, L3 B2B Auto-Procurement product GA. **2027 and Beyond**: Compliance engine covering 20–50+ jurisdictions, developer platform public beta, L4 payment scenario pilots. --- ## 14. Synergy with the LLM Ecosystem ### The Payment Bottleneck in LLM Commercialization LLMs have given Agents the ability to "think," but Agents still cannot "act" — particularly for actions involving fund flows. This bottleneck blocks numerous commercially valuable Agent application scenarios. ### What TuBo Tech Provides to the LLM Ecosystem - **MCP Standard Interfaces**: Any Agent development team can invoke TuBo Tech's payment capabilities through standardized MCP interfaces - **Compliance Training Data**: Real cross-border compliance judgment cases for LLM fine-tuning - **Agent Identity Research Resources**: Real-world Agent identity credentials and behavioral data from multi-market deployments for Agent safety and alignment research - **B2B Commercialization Path**: "Making your Agent capable of payment" is one of the most direct value-adds for LLM teams seeking commercial deployment ### Integration Example An LLM team developing an enterprise customer service Agent needs auto-refund capability. Without TuBo Tech: 3–6 months of development for identity verification, compliance review, and fund flow management. With TuBo Tech: integrate SDK (~1 day), register KYA identity + set boundaries (~1 hour), invoke refund via MCP (~2 hours). **Total integration time: from 3–6 months to 2 days.** --- ## 15. Conclusion We stand at a historic inflection point: AI Agents are about to upgrade from "advisors" to "executors." The key to this upgrade lies not only in payment execution but in compliance, risk control, and cross-border financial service capabilities — only when Agents can transact securely, compliantly, and autonomously can they truly replace humans in end-to-end commercial tasks. The Payment Autonomy Level model provides a clear roadmap for this evolution. The industry is at the L2→L3 inflection point — Visa, Mastercard, Stripe, OpenAI, Coinbase, and UnionPay have all entered the arena. As a pioneer in Agentic Fintech Infrastructure, TuBo Tech provides the full-stack infrastructure for this transition. Through our modular Payment Lego architecture — KYA identity verification, WISE multi-dimensional credit scoring, LLM-driven compliance reasoning and risk control engine, and Vault programmable payment engine — we enable any AI Agent to rapidly gain secure, compliant, and controllable payment and financial service capabilities. In cross-border remittance and B2B settlement scenarios, our compliance reasoning and risk control capabilities provide robust safety guarantees for Agent-driven transactions. Our architecture aligns with both MCP international standards and UnionPay's APOP domestic framework, proven through live deployments across real cross-border payment and remittance scenarios in multiple emerging markets. From L2 to L3, AI Agents will gain the power to "autonomously execute financial services within boundaries" for the first time. TuBo Tech is building this bridge. --- **Contact** - Company: TuBo Tech (Shanghai TuBo Technology Co., Ltd.) - Contact: 吕新浩 (Peter Lyu), CEO & Co-Founder - Email: Peter@tubotech.ai --- *Copyright TuBo Tech. Version 1.0, April 2026.*