https://github.com/agentoverlay/ueta-agent-error-handling-demos
Uniform Electronic Transactions Act based error handling agent demos. Collaboration between Andor Kesselman and Dazza Greenwood
https://github.com/agentoverlay/ueta-agent-error-handling-demos
agentic agentic-ai demo electrontic-payments legal-ai transaction ueta
Last synced: 2 months ago
JSON representation
Uniform Electronic Transactions Act based error handling agent demos. Collaboration between Andor Kesselman and Dazza Greenwood
- Host: GitHub
- URL: https://github.com/agentoverlay/ueta-agent-error-handling-demos
- Owner: agentoverlay
- License: mit
- Created: 2025-03-01T04:37:47.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2025-03-03T07:55:09.000Z (3 months ago)
- Last Synced: 2025-03-09T13:02:08.181Z (2 months ago)
- Topics: agentic, agentic-ai, demo, electrontic-payments, legal-ai, transaction, ueta
- Language: TypeScript
- Homepage:
- Size: 86.9 KB
- Stars: 3
- Watchers: 1
- Forks: 1
- Open Issues: 8
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# UETA Agent -- Goods Producer Demo
This repository showcases how AI-driven agents can integrate with legal frameworks and electronic transactions in the context of a modern goods producer. Inspired by Section 10 of the UETA (Uniform Electronic Transactions Act), the demo features a business API that offers a product catalog and processes orders, while both an autonomous agent (with a dashboard) and a human operator (with a separate dashboard) can interact with transactions. The system emphasizes robust error handling, auditability, and progressive confirmation of transactions—occasionally requiring manual approval (with a 1/10 probability for agent orders) before orders are finalized.
Developed in collaboration between Dazza Greenwood and Andor Kesselman, this project builds on the insights presented in [UETA and LLM Agents: A Deep Dive into Legal Error Handling](https://www.dazzagreenwood.com/p/ueta-and-llm-agents-a-deep-dive-into). In that article, the author explores the challenges and opportunities in merging AI-assisted legal processes with electronic transactions, including transparency, accountability, and the need for clear error correction workflows. This demo reflects those principles, integrating technical and legal nuances to simulate a secure and auditable digital commerce system.
## 📌 Overview
This repository contains demos that illustrate several aspects of AI-assisted legal transactions and goods ordering:
- **Product Catalog and Order Processing**
- The business API acts as a goods producer, providing a catalog of products (SKU, description, and price) and processing orders from users and agents.
- **Error Simulation and Approval**
- An optional error simulation mode (activated with a flag) randomly fails approximately 10% of orders. Additionally, agent orders have a 1/10 chance of requiring manual approval before being finalized.
- **User Wallet and Order Transactions**
- Users (or agents) maintain a digital wallet. Orders deduct funds from the wallet based on product prices.
- **Autonomous Agent with Dashboard**
- An agent mode automatically places random orders and also provides a dashboard to view pending (i.e. pre-approved) transactions.
- **Human Intervention Dashboard**
- A separate human service provides a dashboard where flagged orders are listed, allowing a human operator to approve pending transactions or revert error orders.
## Features
* **Audit Tracing** — If `auditableLog` is set to true, both the producer and consumer maintain logs to mitigate liability.
* **Progressive Confirmation** — When enabled (via config or random chance for agent orders), transactions are held pending approval.
* **Dashboards for Agents and Humans** — Agents can view their pending orders via a dedicated dashboard, while human operators can review and act on flagged transactions.
### Roadmap
* Enhanced Progressive Confirmation
* Improved Detection Mechanisms
* Error Correction Workflows
* System State Tracking
* Support for Two- and Three-Party Agreements
## 🚀 Getting Started
### Prerequisites
[Node.js](https://nodejs.org/) (version 14 or later)
- [pnpm](https://pnpm.io/)
## Demo
**Work in progress**
## Installation
1. Clone the repository.
2. Install dependencies by running:
```bash
pnpm install
```
## Running the Application
### 1. Start the Business API
The business API simulates a goods producer by offering a product catalog and processing orders on port **4000**.
- **Normal mode:**
```bash
pnpm run start:seller
```
- **Error simulation mode:** (Approximately 10% of orders will trigger a simulated error or require approval)
```bash
pnpm run start:seller -- --with-error
```
### 2. Use the Agent CLI & Dashboard
The agent CLI lets you create an account with a wallet, list available products, place orders (with a 1/10 probability that approval is required), and run autonomous agent mode. A dashboard is also available to view pending orders for your account.
#### a. Create an Account
This command creates a new account with a wallet. Each new agent starts with a balance of 1000. You can also deduct an initial deposit (default: 100) if desired.
- **Default deposit:**
```bash
pnpm run start:agent create-account
```
- **Custom deposit (e.g., 200):**
```bash
pnpm run start:agent create-account -- --deposit 200
```
#### b. List Available Products
Display the product catalog from the business API:
```bash
pnpm run start:agent list-products
```
#### c. Place an Order
Submit an order by specifying the product SKU and quantity. Use the `--agent` flag to indicate the order is placed by the agent. For example, to order 2 units of the product with SKU `SKU001`:
```bash
pnpm run start:agent order -- --sku SKU001 --quantity 2 --agent
```
#### d. Start Autonomous Agent Mode
Launch autonomous agent mode, which automatically places random orders at random intervals:
```bash
pnpm run start:agent agent
```
#### e. Launch the Agent Dashboard
View pending (pre-approved) orders for your account via a dashboard. By default, the dashboard runs on port **6001**:
```bash
pnpm run start:agent dashboard
```
### 3. Use the Human Intervention Dashboard
The human service provides a dashboard (accessible via a web browser) where flagged orders are listed. Here, a human operator can:
- **Approve** pending transactions (moving them from `pending_confirmation` to `delivered`).
- **Revert** orders that encountered errors.
The human service listens on a configurable port (default **5002**). To change it, set the environment variable `HUMAN_PORT`.
To start the human service:
```bash
pnpm run start:human
```
Then, access the dashboard at:
```
http://localhost:5002/dashboard
```
### 4. Viewing Order Logs
Logs for both the business and agent sides are maintained (if enabled via `auditableLog`) to support audit tracing and error correction.
## Project Structure
- **business.ts**
The business API server simulates a goods producer. It maintains a product catalog, processes orders (with optional error simulation and approval requirements), and logs transaction statuses.
- **agent.ts**
The agent CLI tool allows account creation, wallet management, product listing, order placement, autonomous order placement, and includes a dashboard for viewing pending orders.
- **human.ts**
The human service provides a dashboard for reviewing flagged orders and supports approving or reverting transactions.
- **config.ts**
Contains configuration settings that control logging, progressive confirmation, and monitoring.
- **package.json**
Contains project scripts and dependencies.
## License
This project is provided as-is under the MIT License.