AgentBL

From recommending products to recommending the assets behind global commerce.
从推荐商品，到推荐支撑全球商业流通的真实资产。

AgentBL is an AI-powered, context-aware platform for evaluating, pricing, recommending, and continuously monitoring electronic Bill of Lading (eBL)-backed Real-World Assets (RWAs).

AgentBL explores how the same intelligent-agent principles used in next-generation conversational e-commerce — intent understanding, retrieval, ranking, multi-turn context, dynamic recommendation, and continuous monitoring — can be extended from consumer products to real-world trade assets.

AgentBL 是一个由 AI 驱动、具备上下文感知能力的平台，用于评估、定价、推荐和持续监控由电子提单（Electronic Bill of Lading, eBL）支持的现实世界资产（Real-World Assets, RWA）。

AgentBL 探索如何将下一代对话式电子商务中的核心智能 Agent 能力——意图理解、检索、排序、多轮上下文、动态推荐与持续监控——从消费商品进一步扩展到现实世界贸易资产。

AI understands the context. AgentBL prices the risk. Web3 coordinates the asset.
AI 理解上下文，AgentBL 为风险定价，Web3 协调现实资产。

🌐 Project Overview | 项目概述

Modern e-commerce is moving beyond static keyword search.

A next-generation shopping agent must understand what a user actually wants, distinguish high-intent Buying from open-ended Browsing, retrieve relevant candidates, rank them according to evolving preferences, and continuously adapt as new context becomes available.

AgentBL applies this same underlying agentic decision pattern to another layer of digital commerce:

Investment in eBL-backed real-world trade assets.

In a conventional e-commerce platform, an AI shopping agent learns a consumer's preferences and recommends suitable products. After a purchase, the commerce and logistics infrastructure continues to track the physical movement of those goods.

AgentBL extends this logic from:

Consumers + Products

to:

Investors + Tokenized Trade Assets

Instead of asking:

Which product best matches this consumer?

AgentBL explores:

Which eBL-backed trade asset best matches this investor's risk and return preferences?

中文

现代电子商务正在逐渐摆脱静态关键词搜索。

下一代 Shopping Agent 需要理解用户真正想要什么，区分具有明确购买意图的 Buying（购买）和开放式探索的 Browsing（浏览），从商品目录中检索候选商品，根据不断变化的用户偏好进行排序，并随着新的 Context 出现持续调整推荐。

AgentBL 将这种底层的 Agentic Decision Pattern（智能体决策模式）应用到数字商业的另一个层面：

基于电子提单的现实世界贸易资产投资。

在传统电子商务平台中，AI Shopping Agent 根据消费者的需求和偏好推荐商品。购买完成之后，电商与物流系统还会继续追踪这些商品在现实世界中的运输状态。

AgentBL 将这一逻辑从：

消费者 + 商品

扩展为：

投资者 + 代币化贸易资产

传统 Shopping Agent 回答的是：

哪个商品最符合这个消费者的需求？

AgentBL 探索的是：

哪个由电子提单支持的贸易资产，最符合这个投资者的风险与收益偏好？

推荐对象发生了变化，但两者共享高度相关的底层智能决策逻辑。

🛍️ From Product Recommendation to Trade-Asset Recommendation

从商品推荐到贸易资产推荐

Conversational E-commerce	AgentBL
Consumer 消费者	Investor 投资者
Product catalog 商品目录	eBL-backed RWA opportunity pool
Consumer preferences 消费者偏好	Investor risk / return preferences 投资者风险/收益偏好
Buying intent 明确购买意图	Targeted investment intent 明确投资意图
Browsing intent 浏览意图	Investment opportunity exploration 投资机会探索
Product attributes 商品属性	Cargo, voyage, maturity, yield and risk attributes
Product retrieval 商品检索	Trade-asset opportunity retrieval 贸易资产检索
Product ranking 商品排序	Risk-adjusted investment ranking 风险调整后的投资排序
Product recommendation 商品推荐	eBL-backed RWA recommendation
Product price 商品价格	RWA valuation / financing price
Logistics tracking 物流追踪	Cargo and voyage risk tracking
Updated user context 用户上下文更新	Updated investor + trade-risk context
Product re-ranking 商品重新排序	Repricing / reassessment 重新定价与评估
Purchase conversion 完成购买	Investment decision 投资决策

Conversational E-commerce | 对话式电商

Consumer Intent / 消费者意图
          ↓
Candidate Retrieval / 候选商品检索
          ↓
Ranking / 排序
          ↓
Product Recommendation / 商品推荐
          ↓
Purchase + Logistics / 购买与物流
          ↓
Updated Context / 上下文更新
          ↓
Re-ranking / 重新排序

AgentBL

Investor Intent / 投资者意图
          ↓
Trade-Asset Retrieval / 贸易资产检索
          ↓
Risk Analysis + Valuation / 风险分析与估值
          ↓
Investment Recommendation / 投资推荐
          ↓
Cargo + Voyage Monitoring / 货物与航程监控
          ↓
Updated Risk Context / 风险上下文更新
          ↓
Revaluation + Re-ranking / 重新估值与排序

The key connection is Dynamic Context.

A consumer's preferred product may change when their requirements change. Likewise, an eBL-backed investment opportunity may become more or less attractive when the real-world shipment changes.

New information about weather, shipping routes, geopolitical events, insurance coverage, cargo conditions, commodity markets, or counterparty risk can become new context for the AgentBL agent.

核心连接点是 Dynamic Context（动态上下文）。

消费者的需求发生变化时，最适合他的商品可能发生变化。

同样，一笔 eBL-backed 投资机会是否值得投资，也会随着现实世界货物状态的变化而变化。

新的天气、航线、地缘政治、保险、货物状态、大宗商品市场以及交易对手风险信息，都可以成为 Agent 的新 Context。

因此，两种系统都可以形成持续决策循环：

Observe / 观察
      ↓
Understand Context / 理解上下文
      ↓
Retrieve / 检索
      ↓
Evaluate / 评估
      ↓
Recommend / 推荐
      ↓
Monitor / 监控
      ↓
Update Context / 更新上下文
      ↓
Re-evaluate / 重新评估

Recommendation is a continuous decision process, not a one-time prediction.
推荐是一个持续运行的决策过程，而不是一次性的预测。

🔄 Dynamic Context Programming | 动态上下文编程

Consider a shopping agent.

A consumer may initially say:

"I need a pair of shoes."

Then:

"They should be waterproof."

Then:

"My budget is below $100."

The agent should accumulate these constraints and continuously re-rank products.

If the consumer suddenly says:

"Actually, I want a jacket."

the agent must recognize an Intent Override and update or erase obsolete state.

中文

例如消费者最开始说：

“我需要一双鞋。”

随后增加条件：

“最好防水。”

然后继续说：

“预算低于 100 美元。”

Shopping Agent 应该持续积累这些条件，并重新 Ranking 商品。

如果消费者突然说：

“算了，我想买一件夹克。”

Agent 则需要识别 Intent Override（意图覆盖），删除已经失效的状态并重建新的 Context。

AgentBL 中同样存在这一逻辑。

投资者的 Context 可能包含：

Risk tolerance / 风险承受能力
Target return / 目标收益
Maturity / 到期期限
Cargo preference / 货物偏好
Geographic exposure / 地理风险敞口
Trade route / 航线
Counterparty risk / 交易对手风险
Insurance conditions / 保险情况

与此同时，现实世界的货物 Context 也会持续变化。

例如：

Typhoon Detected / 发现台风
          ↓
Voyage Delay Probability ↑ / 航程延误概率上升
          ↓
Cargo + Insurance Risk Changes / 货物与保险风险变化
          ↓
Agent Context Updates / Agent 更新上下文
          ↓
Risk Reassessment / 重新评估风险
          ↓
Asset Repricing / 资产重新定价
          ↓
Recommendation Changes / 投资推荐改变

When real-world context changes, the recommendation should change with it.
当现实世界的 Context 发生变化时，Recommendation 也应该随之改变。

🌍 From Web2 Commerce to Web3 Trade Infrastructure

从 Web2 电商到 Web3 贸易基础设施

E-commerce does not end when a customer clicks Buy.

Behind every digital transaction is a physical supply chain in which goods must be documented, transported, tracked, financed, transferred, and settled.

The electronic Bill of Lading (eBL) is an important digital trade document associated with the physical movement and control of goods.

AgentBL explores how eBL-backed trade flows can connect with Real-World Asset (RWA) infrastructure.

电子商务并不会在消费者点击 Buy 的那一刻结束。

数字交易背后仍然存在完整的现实世界供应链，商品需要经历：

Document / 单据
      ↓
Transport / 运输
      ↓
Track / 追踪
      ↓
Finance / 融资
      ↓
Transfer / 转移
      ↓
Settle / 结算

电子提单（eBL）是连接数字商业与现实世界货物流转的重要数字贸易文件。

AgentBL 探索如何进一步将 eBL 所代表的现实贸易流程与 RWA 基础设施连接起来。

从而形成更完整的商业生命周期：

Discover / 发现
      ↓
Recommend / 推荐
      ↓
Buy / 购买
      ↓
Ship / 运输
      ↓
Track / 追踪
      ↓
Finance / 融资
      ↓
Settle / 结算

In Web2 e-commerce:

Understand the consumer → Recommend the product

AgentBL extends this idea:

Understand the investor → Evaluate the trade asset → Monitor the cargo → Update risk → Reprice → Support financing and settlement

在 Web2 电商中：

理解消费者 → 推荐商品

AgentBL 则进一步探索：

理解投资者 → 评估贸易资产 → 监控现实货物 → 更新风险 → 重新定价 → 支持融资与结算

The broader vision is an Intelligent Commerce Infrastructure in which:

consumers receive personalized product recommendations;
investors receive context-aware trade-asset recommendations;
real-world cargo information continuously updates the decision context;
changing risk triggers reassessment and repricing;
eBL-backed RWAs connect physical trade with programmable financing;
Web2 commerce workflows can interact with Web3 financial infrastructure.

我们的长期目标是一种 Intelligent Commerce Infrastructure（智能商业基础设施）：

消费者获得更加个性化的商品推荐；
投资者获得符合自身 Context 的贸易资产推荐；
现实世界货物信息持续进入 Agent 的决策 Context；
风险变化能够触发重新评估和重新定价；
eBL-backed RWA 将现实贸易与可编程融资连接；
Web2 商业工作流逐渐与 Web3 金融基础设施交互。

🎯 Alignment with the Conversational Commerce Challenge

与 Conversational Commerce Challenge 的对应关系

The challenge focuses on intelligent recommendation under evolving conversational context.

AgentBL explores how the same architectural principles can be generalized from Product Recommendation to Trade-Asset Recommendation.

本次比赛关注的是：

如何在不断变化的对话 Context 中进行高质量智能推荐。

AgentBL 探索如何将相同的 Agent Architecture 从：

商品推荐

进一步推广到：

贸易资产推荐

I. Intent Routing & Hybrid Decision Pipeline

意图路由与混合决策 Pipeline

The challenge distinguishes between:

Buying
Browsing

A similar distinction exists in investment discovery.

比赛要求区分：

Buying：具有明确目标的购买
Browsing：开放式浏览

投资场景中也存在类似区别。

Targeted Investment Intent | 明确投资意图

An investor may specify:

acceptable risk;
target yield;
maturity;
cargo type;
geographic exposure;
trade route.

投资者可能明确提出：

可接受风险；
目标收益；
到期期限；
货物类别；
地理风险；
航线。

这些属于 Hard Constraints。

Exploratory Investment Intent | 探索式投资意图

An investor may instead ask:

"What trade-finance opportunities are worth looking at?"

投资者也可能只问：

“最近有什么值得关注的贸易融资机会？”

这种行为类似于 Browsing，需要更广泛地发现候选机会。

共同 Pipeline：

Intent Detection / 意图识别
          ↓
Candidate Retrieval / 候选检索
          ↓
Semantic + Risk Ranking / 语义与风险排序
          ↓
Recommendation / 推荐

II. Multi-Turn Scenario Evolution

多轮场景演化

Shopping and investment decisions both evolve over time.

Users may progressively reveal preferences or abruptly change their requirements.

购物和投资决策都会随着对话不断变化。

Agent 应维护不断变化的 State，而不是把每次输入当作一个完全独立的 Query。

Relevant state may include:

risk tolerance;
expected return;
maturity;
commodity preference;
geographic exposure;
voyage conditions;
counterparty risk;
insurance conditions.

对应状态可能包括：

风险承受能力；
目标收益；
到期期限；
商品/货物类别偏好；
地理风险；
航程状况；
交易对手风险；
保险情况。

III. Dynamic Context Programming

动态上下文编程

For conversational shopping:

New User Preference
        ↓
Context Update
        ↓
Product Re-ranking

对于 AgentBL：

New Cargo / Voyage / Risk Evidence
              ↓
        Context Update
              ↓
       Risk Reassessment
              ↓
           Repricing
              ↓
   Investment Re-ranking

因此，极端天气、地缘政治冲突、保险问题或交易对手风险的变化，都可能在 Runtime 改变一项投资机会的吸引力。

IV. Recommendation Quality & Decision Efficiency

📌 Scope of This Repository | 本仓库范围

AgentBL was originally developed around:

AI-assisted eBL-backed RWA trade finance

with components covering:

risk analysis;
pricing;
paid intelligence;
MCP interoperability;
Injective-based execution.

AgentBL 最初围绕：

AI-assisted eBL-backed RWA Trade Finance

进行开发，目前包含：

风险分析；
AI 定价；
Paid Intelligence；
MCP 互操作；
Injective 链上执行。

For the conversational-commerce challenge, AgentBL demonstrates how the same underlying agent architecture can be applied to recommendation under dynamic context.

The competition's Amazon catalog and official evaluation framework remain a separate, read-only evaluation environment.

AgentBL does not claim that Amazon products themselves are tokenized as RWAs.

对于本次 Conversational Commerce Challenge，我们探索的是相同 Agent Architecture 在 Dynamic Context Recommendation 中的应用。

比赛提供的 Amazon Catalog 与官方 Evaluation Framework 仍然属于一个独立、只读的评估环境。

AgentBL 不会声称 Amazon 商品本身被 Tokenize 成 RWA。

真正连接两者的是底层智能决策架构：

Preference Understanding + Context
                ↓
             Retrieval
                ↓
              Ranking
                ↓
          Recommendation
                ↓
             Monitoring
                ↓
          Context Update
                ↓
           Re-evaluation

即：

理解偏好与上下文 → 检索 → 排序 → 推荐 → 监控 → 更新上下文 → 重新评估

🧠 System Architecture | 系统架构

AgentBL combines four major layers:

┌──────────────────────────────────────────────┐
│              AI / Agent Layer                │
│ Risk Analysis • Pricing • Recommendation    │
│ Dynamic Context • Evidence Evaluation       │
└──────────────────────┬───────────────────────┘
                       ↓
┌──────────────────────────────────────────────┐
│           x402 Intelligence Layer            │
│ Paid Risk Reports • Valuation • Settlement  │
└──────────────────────┬───────────────────────┘
                       ↓
┌──────────────────────────────────────────────┐
│              MCP Interoperability            │
│ Agent Tools • Resources • External AI       │
└──────────────────────┬───────────────────────┘
                       ↓
┌──────────────────────────────────────────────┐
│          Web3 / Injective Execution          │
│ eBL • RWA • Pricing Oracle • Settlement     │
└──────────────────────────────────────────────┘

Architecture Color Legend | 颜色说明

🤖 Purple / 紫色 — AI / ML 与自主决策
💰 Orange / 橙色 — x402 付费情报与收入层
⛓️ Blue / 蓝色 — Web3 / Injective 结算与执行
🔌 Teal / 青绿色 — MCP 互操作层

🤖 AI Risk & Recommendation Engine

AI 风险与推荐引擎

AgentBL evaluates trade opportunities using multiple sources of context:

cargo value / 货物价值
trade documentation / 贸易文件
voyage information / 航程信息
insurance / 保险
geopolitical conditions / 地缘政治
weather / 天气
counterparty risk / 交易对手风险
collateral safety / 抵押品安全性
market conditions / 市场环境

The objective is not simply to produce a risk score.

目标并不仅仅是产生一个 Risk Score，而是形成：

Evidence / 证据
      ↓
Risk Assessment / 风险分析
      ↓
Valuation / 估值
      ↓
Recommendation / 推荐
      ↓
Monitoring / 监控
      ↓
Reassessment / 重新评估

A deterministic fallback remains available when an external LLM is unavailable in Demo Mode.

在 Demo Mode 下，即使外部 LLM 不可用，系统仍然提供 Deterministic Fallback。

🚢 Cargo Risk as Dynamic Recommendation Context

将货物风险作为动态推荐上下文

Logistics information is not merely something displayed after an investment.

It is part of the decision context.

物流信息并不是投资完成后才展示的附属信息，而是 Recommendation Context 的组成部分。

Normal Voyage
→ Stable Risk
→ Offering Remains Attractive

但如果发生：

Typhoon / War / Insurance Issue
        ↓
Risk Context Changes
        ↓
AI Reassesses Exposure
        ↓
Pricing Changes
        ↓
Recommendation May Change

This closely parallels conversational e-commerce, where new consumer information triggers product re-ranking.

这与 Conversational E-commerce 中：

新的消费者偏好 → 商品重新 Ranking

具有相似的动态决策结构。

💰 x402 Paid Intelligence | x402 付费情报

AgentBL includes x402-based paid intelligence flows.

Machine clients and AI agents can request premium risk or valuation information through a payment-gated interface.

AgentBL 包含基于 x402 的 Paid Intelligence Flow。

机器客户端和其他 AI Agent 可以通过支付接口获取高级风险和估值信息。

AI Agent
      ↓
Requests Risk Intelligence / 请求风险情报
      ↓
HTTP 402
      ↓
Payment / Authorization
      ↓
Settlement
      ↓
Premium Intelligence Unlocked

Demo Mode can reproduce the core flow without requiring a live wallet.

Demo Mode 下无需真实钱包即可演示核心流程。

🔌 MCP Interoperability | MCP 互操作

AgentBL exposes capabilities through the Model Context Protocol (MCP).

The repository includes MCP tools and resources for operations such as:

retrieving trade cases;
generating pricing quotes;
searching the knowledge base;
simulating offerings;
pushing pricing information to the oracle.

AgentBL 通过 Model Context Protocol（MCP）向其他 AI 系统提供结构化能力，包括：

获取 Trade Case；
生成 Pricing Quote；
搜索 Knowledge Base；
模拟 Offering；
向 Oracle 推送 Pricing 信息。

This allows external AI systems to interact with AgentBL as a structured tool provider rather than relying only on a web interface.

这意味着其他 AI 系统可以直接把 AgentBL 作为 Tool Provider，而不必依赖网页界面。

⛓️ Web3 & Injective

AgentBL includes smart contracts for representing and coordinating the eBL-backed RWA lifecycle.

AgentBL 包含用于表示和协调 eBL-backed RWA 生命周期的 Smart Contracts，包括：

eBL registration / eBL 登记
RWA offerings / RWA 发行
risk pricing / 风险定价
tokenized RWA shares / RWA Token 份额
protocol coordination / 协议协调

The architecture uses Injective's EVM-compatible environment for Web3 execution.

项目使用 Injective 的 EVM-compatible 环境进行 Web3 执行。

📍 On-Chain Evidence | 链上证据

Deployed Contracts — Injective Testnet

已部署合约 — Injective 测试网

Chain ID: 1439

Contract 合约	Address 地址	Purpose 用途
AgentBLRWA	`0x3C6774d2D1ac6Bf35d08e5C9c84A233F5bc7f5Ce`	Protocol coordinator / 协议协调
EBLRegistry V2	`0x85bfdcd00E0bBb9dDce3dcD2A58A62380703AdA6`	eBL registry / anti-duplication
RWAOfferingPool	`0x1F44D336111dE4e2640bd9a5945991D42e876f65`	Offering lifecycle / 生命周期
RiskPricingOracle	`0x0F9618DDbac86eD51d48ef1361789D7e5eF1FAE1`	AI → on-chain pricing
RWAToken	`0x7eb44f73368d14DBE4c2E30F8490a60513Fe17B0`	ERC-20 RWA shares

🧪 Reproducible Verification | 可复现验证

Requirements | 环境要求

Node.js 20+
npm
Git

Install | 安装

npm install

Preflight

npm run preflight

The preflight gate checks:

runtime environment;
repository integrity;
Node tests;
Solidity contracts;
API flows;
x402;
MCP;
UI acceptance;
live-readiness configuration.

Preflight 会检查：

Runtime Environment；
Repository Integrity；
Node Tests；
Solidity Contracts；
API Flow；
x402；
MCP；
UI；
Live Readiness。

Live-network checks may be skipped or reported as warnings in Demo Mode.

Demo Mode 下，Live Network 检查可能被跳过或显示 WARN。

Solidity Contract Tests | Solidity 合约测试

cd hardhat
npm install
npm test

A locally reproduced run produced:

32 passing

本地复现结果：

32 passing

然后返回项目根目录：

cd ..

🚀 Quick Start | 快速开始

🎯 Judge / Demo Mode | 评委快速验证

git clone https://github.com/Sunnypkulaw/Agent-BL.git
cd Agent-BL

npm install

cd hardhat
npm install
cd ..

npm run preflight
npm run dev

Open / 打开：

http://localhost:3000

Demo Mode does not require a live wallet for the core demonstration.

Demo Mode 的核心演示不需要真实钱包。

What You Can Explore | 可以体验

RWA investment opportunities / RWA 投资机会
AI risk assessments / AI 风险评估
pricing and expected yield / 定价与预期收益
cargo and voyage risk / 货物与航程风险
investment workflow / 投资流程
x402 paid intelligence / x402 付费情报
MCP integration / MCP 集成
Web3 evidence / Web3 链上证据

👨‍💻 Developer Mode | 开发者模式

Install dependencies:

npm install

Install Hardhat dependencies:

cd hardhat
npm install
cd ..

Optional environment configuration:

cp .env.example .env

Never commit .env, private keys, wallet secrets, or API keys to GitHub.
绝对不要把 .env、Private Key、钱包 Secret 或 API Key 提交到 GitHub。

Useful commands / 常用命令：

npm run dev
npm run preflight
npm test
npm run check
npm run demo
npm run scenarios
npm run price
npm run intel
npm run smoke:x402
npm run mcp:stdio

Contract tests / 合约测试：

npm --prefix hardhat test

🧪 Dynamic Risk Scenarios | 动态风险场景

Normal Trade | 正常贸易

Normal Shipment / 正常运输
→ Acceptable Risk / 风险可接受
→ Financing Remains Available / 融资继续开放

Severe Weather | 极端天气

Typhoon / 台风
→ Voyage Disruption / 航程受到影响
→ Increased Risk / 风险上升
→ Repricing / Possible Pause / 重新定价或暂停

Geopolitical Crisis | 地缘政治危机

Conflict / 冲突
→ Route + Insurance + Counterparty Uncertainty
→ Increased Risk
→ Recommendation Reassessed

Documentation / Fraud Risk | 文件与欺诈风险

Document Inconsistency / 文件不一致
→ Evidence Validation Fails
→ Risk Increases
→ Offering Rejected or Paused

These scenarios demonstrate why Dynamic Context matters for trade-asset recommendation.

这些场景展示了为什么 Dynamic Context 对贸易资产 Recommendation 至关重要。

🛠️ Technology Stack | 技术栈

AI & Intelligence

Context-aware risk assessment
Deterministic fallback
RAG-style knowledge retrieval
Structured output with Zod
Scenario-based risk evaluation

Blockchain & Web3

Injective EVM-compatible testnet
Solidity
Hardhat
ethers.js 6.x

Payment & Settlement

@injectivelabs/x402
x402 payment-gated intelligence
Settlement state machine
Payment / report evidence binding

Backend

Node.js 20+
Express 5.x
Native Node.js test runner
JSON-based local state

Agent Interoperability

@modelcontextprotocol/sdk
MCP tools
MCP resources
Standalone MCP server

Frontend

Vanilla JavaScript
HTML
Modern CSS
Interactive AgentBL Dashboard

💼 Business Model | 商业模式

AgentBL explores several potential revenue channels.

AgentBL 探索多个潜在收入方向。

1. Paid Risk Intelligence | 付费风险情报

Risk, fraud, and valuation reports can be provided as machine-consumable intelligence.

风险、欺诈与估值报告可以作为 Machine-readable Intelligence 提供给：

Investors / 投资者
Banks / 银行
Insurers / 保险公司
Logistics platforms / 物流平台
Autonomous AI agents / 自主 AI Agent

2. RWA Protocol Fees | RWA 协议费用

A future production implementation could charge protocol fees on successfully financed trade assets.

未来 Production Version 可以对成功融资的贸易资产收取协议费用。

3. Institutional APIs | 机构 API

Banks, insurers, and trade-finance platforms could integrate AgentBL through APIs or MCP-compatible workflows.

银行、保险公司和贸易融资平台可以通过 API 或 MCP-compatible Workflow 接入 AgentBL。

AgentBL is currently a hackathon prototype, not a production financial service.
AgentBL 当前是 Hackathon Prototype，而不是正式金融服务。

⚠️ Limitations | 当前局限

AgentBL is currently a hackathon prototype.

Important limitations include:

cargo and settlement scenarios may use simulated or demo data;
Demo Mode does not represent real investment execution;
real-world legal ownership requires jurisdiction-specific integration;
production eBL infrastructure requires recognized eBL systems and standards;
real logistics feeds are not yet fully integrated;
production insurance and commodity data require reliable external providers;
AI risk assessment requires human and institutional controls before real financial use;
this repository is not investment advice;
the competition's Amazon retrieval benchmark remains separate from the current AgentBL RWA demonstration.

AgentBL 当前仍然是一个 Hackathon Prototype。

主要限制包括：

部分货物与结算场景使用模拟 / Demo 数据；
Demo Mode 不代表真实投资执行；
现实世界法律所有权需要针对司法辖区进行进一步集成；
Production eBL 需要接入被认可的 eBL 系统与行业标准；
真实物流数据尚未完整接入；
保险与商品市场数据需要 Production-grade Data Provider；
AI 风险评估在真实金融环境中需要人工和机构控制；
本项目不构成投资建议；
比赛的 Amazon Retrieval Benchmark 与当前 AgentBL RWA Demo 属于不同的验证环境。

🔮 What We Would Improve With More Time

如果有更多时间，我们会改进什么？

Conversational Commerce

integrate the official frozen Amazon competition catalog
接入官方冻结的 Amazon Catalog
implement explicit Buying / Browsing intent routing
实现明确的 Buying / Browsing Intent Routing
add keyword + category + dense multi-route retrieval
加入 Keyword + Category + Dense Multi-route Retrieval
add LLM semantic re-ranking
加入 LLM Semantic Re-ranking
implement session-level preference distillation
实现 Session-level Preference Distillation
support slot accumulation, decay, erasure, and intent override
支持 Slot Accumulation、Decay、Erasure 与 Intent Override
add proactive clarification when the candidate pool remains too broad
Candidate Pool 过大时主动提出 Clarification
evaluate Hit Rate@K, MRR, Top-K Hit Rate, and MTTC
使用官方 Evaluator 测试 Hit Rate@K、MRR、Top-K Hit Rate 与 MTTC

AgentBL / Trade Finance

integrate real-time logistics data
接入实时物流数据
integrate production-grade insurance and commodity risk feeds
接入 Production-grade 保险与大宗商品风险数据
strengthen investor preference modelling
加强投资者 Preference Modelling
dynamically rank multiple RWA opportunities
对多个 RWA Opportunity 进行 Dynamic Ranking
connect cargo events directly to recommendation updates
将 Cargo Event 直接连接 Recommendation Update
strengthen KYC / AML and investor eligibility controls
加强 KYC / AML 与投资者资格控制
extend eBL legal interoperability
加强 eBL 法律互操作性
improve AI explainability
提高 AI 风险与 Recommendation 的可解释性

The longer-term goal is to connect:

长期目标是连接：

Conversational Commerce
        +
Dynamic Recommendation
        +
Real-World Logistics
        +
Trade Finance
        +
Web3 Settlement

🗺️ Roadmap | 路线图

Current Prototype | 当前 Prototype

AI-assisted trade risk and pricing
eBL-backed RWA architecture
Investment Opportunity Dashboard
Cargo / Voyage Risk Scenarios
x402 Paid Intelligence
MCP Interoperability
Solidity Smart Contracts
Injective Testnet Evidence
Local Demo Mode

Next Stage | 下一阶段

Conversational Investor Preference Modelling
Dynamic Opportunity Ranking
Richer Logistics Context
Real-world Data Integration
Stronger Recommendation Explainability
Competition-compatible Product Retrieval

Long-Term Vision | 长期愿景

AI understands demand, context, risk and opportunity, while programmable financial infrastructure coordinates the assets behind real-world commerce.

AI 理解需求、上下文、风险与机会，而可编程金融基础设施负责协调现实商业背后的资产。

Product Discovery / 商品发现
          ↓
Recommendation / 推荐
          ↓
Purchase / 购买
          ↓
Physical Trade / 现实贸易
          ↓
eBL / 电子提单
          ↓
RWA Financing / RWA 融资
          ↓
Continuous Risk Monitoring / 持续风险监控
          ↓
Programmable Settlement / 可编程结算

❓ FAQ | 常见问题

Q1. Is AgentBL a shopping recommendation engine?

AgentBL 是 Shopping Recommendation Engine 吗？

Not directly.

AgentBL currently focuses on trade-asset evaluation and recommendation, rather than Amazon product recommendation.

However, both share the same underlying decision pattern:

目前不是直接意义上的 Amazon 商品推荐引擎。

AgentBL 当前重点是贸易资产评估与推荐，但两者共享类似的底层 Agent Decision Pattern：

Understand Preference / 理解偏好
        ↓
Retrieve Candidates / 检索候选
        ↓
Evaluate / 评估
        ↓
Rank / 排序
        ↓
Recommend / 推荐
        ↓
Monitor Context / 监控上下文
        ↓
Re-evaluate / 重新评估

Q2. Why is AgentBL relevant to conversational e-commerce?

为什么 AgentBL 与对话式电商有关？

Because both systems address the same broader problem:

How can an intelligent agent turn changing, incomplete human intent and external context into a high-quality recommendation?

因为两者都在解决同一个更广义的问题：

智能 Agent 如何把不断变化、不完整的人类意图和外部 Context 转化成高质量 Recommendation？

A shopping agent uses:

Consumer Preferences + Product Information

AgentBL uses:

Investor Preferences + Trade Information + Dynamic Cargo Risk

Q3. Why use an electronic Bill of Lading?

为什么使用电子提单？

An eBL creates a digital bridge between:

eBL 在以下环节之间建立数字化连接：

Digital Commerce / 数字商业
        ↓
Physical Goods / 现实货物
        ↓
Trade Documentation / 贸易文件
        ↓
Financing / 融资

AgentBL explores how this bridge can support eBL-backed RWA workflows.

AgentBL 进一步探索如何利用这一连接建立 eBL-backed RWA Workflow。

Q4. What happens when cargo risk changes?

如果货物风险发生变化怎么办？

New Evidence / 新风险信息
        ↓
Risk Context Changes / 风险上下文变化
        ↓
AI Reassesses Trade / AI 重新评估
        ↓
Pricing Changes / 定价变化
        ↓
Investment Recommendation Changes / 投资推荐变化

This is analogous to a shopping agent re-ranking products after consumer preferences change.

这相当于 Shopping Agent 在消费者偏好变化之后重新 Ranking 商品。

Q5. Is AgentBL a production investment platform?

AgentBL 是正式投资平台吗？

No.

AgentBL is currently a hackathon prototype demonstrating the combination of:

AI
Dynamic Context
Risk Assessment
Web3
eBL / RWA
x402
MCP
Trade Finance

AgentBL 当前是一个 Hackathon Prototype，用于展示 AI、Dynamic Context、风险分析、Web3、eBL/RWA、x402、MCP 与 Trade Finance 的结合。

It is not investment advice and should not be treated as a production financial service.
本项目不构成投资建议，也不应被视为正式金融服务。

📄 License

MIT License — see LICENSE.

MIT License — 详见 LICENSE。

### From product recommendation to the assets behind global commerce. ### 从商品推荐，到支撑全球商业流通的真实资产。 **AI understands the context. AgentBL prices the risk. Web3 coordinates the asset.** **AI 理解上下文，AgentBL 为风险定价，Web3 协调现实资产。**

Built With

azure-ai-foundry
azure-openai
blockchain
ethers.js
gpt-4o
iagent-framework
inevm
injective
mcp-server
node.js
python
rest
rwa
smart-contracts
solidity
typescript
usdc
web3
x402

AgentBL

🌐 Project Overview | 项目概述

中文

🛍️ From Product Recommendation to Trade-Asset Recommendation

从商品推荐到贸易资产推荐

Conversational E-commerce | 对话式电商

AgentBL

🔄 Dynamic Context Programming | 动态上下文编程

中文

🌍 From Web2 Commerce to Web3 Trade Infrastructure

从 Web2 电商到 Web3 贸易基础设施

🎯 Alignment with the Conversational Commerce Challenge

与 Conversational Commerce Challenge 的对应关系

I. Intent Routing & Hybrid Decision Pipeline

意图路由与混合决策 Pipeline

Targeted Investment Intent | 明确投资意图

Exploratory Investment Intent | 探索式投资意图

II. Multi-Turn Scenario Evolution

多轮场景演化

III. Dynamic Context Programming

动态上下文编程

IV. Recommendation Quality & Decision Efficiency

推荐质量与决策效率

📌 Scope of This Repository | 本仓库范围

🧠 System Architecture | 系统架构

Architecture Color Legend | 颜色说明

🤖 AI Risk & Recommendation Engine

AI 风险与推荐引擎

🚢 Cargo Risk as Dynamic Recommendation Context

将货物风险作为动态推荐上下文

💰 x402 Paid Intelligence | x402 付费情报

🔌 MCP Interoperability | MCP 互操作

⛓️ Web3 & Injective

📍 On-Chain Evidence | 链上证据

Deployed Contracts — Injective Testnet

已部署合约 — Injective 测试网

🧪 Reproducible Verification | 可复现验证

Requirements | 环境要求

Install | 安装

Preflight

Solidity Contract Tests | Solidity 合约测试

🚀 Quick Start | 快速开始

🎯 Judge / Demo Mode | 评委快速验证

What You Can Explore | 可以体验

👨‍💻 Developer Mode | 开发者模式

🧪 Dynamic Risk Scenarios | 动态风险场景

Normal Trade | 正常贸易

Severe Weather | 极端天气

Geopolitical Crisis | 地缘政治危机

Documentation / Fraud Risk | 文件与欺诈风险

🛠️ Technology Stack | 技术栈

AI & Intelligence

Blockchain & Web3

Payment & Settlement

Backend

Agent Interoperability

Frontend

💼 Business Model | 商业模式

1. Paid Risk Intelligence | 付费风险情报

2. RWA Protocol Fees | RWA 协议费用

3. Institutional APIs | 机构 API

⚠️ Limitations | 当前局限

🔮 What We Would Improve With More Time

如果有更多时间，我们会改进什么？

Conversational Commerce

AgentBL / Trade Finance

🗺️ Roadmap | 路线图

Current Prototype | 当前 Prototype

Next Stage | 下一阶段

Long-Term Vision | 长期愿景

❓ FAQ | 常见问题

Q1. Is AgentBL a shopping recommendation engine?

AgentBL 是 Shopping Recommendation Engine 吗？

Q2. Why is AgentBL relevant to conversational e-commerce?

为什么 AgentBL 与对话式电商有关？

Q3. Why use an electronic Bill of Lading?

为什么使用电子提单？

Q4. What happens when cargo risk changes?

如果货物风险发生变化怎么办？

Q5. Is AgentBL a production investment platform?