Project: Mem U Trader

※発表内容、demo動画は/slide /movieにあります！

Overview

This project is a sample application demonstrating the integration of MemU (an agentic memory layer for LLM applications). It consists of a React-based frontend and a Flask-based backend that acts as an interface to the MemU API.

Tech Stack

Frontend

• Framework: React 19 (via Vite)
• Language: TypeScript
• Styling: Tailwind CSS, PostCSS
• UI Components: Radix UI (via shadcn/ui patterns), Lucide React
• Build Tool: Vite

Backend

• Framework: Flask (Python)
• Dependencies: flask-cors, requests, openai
• External Service: MemU API (v3), OpenAI API

Directory Structure

memu_sample/
├── frontend/           # React application
│   ├── src/
│   │   ├── components/ # UI components (Button, Input, etc.)
│   │   ├── lib/        # Utilities
│   │   ├── App.tsx     # Main application component
│   │   └── main.tsx    # Entry point
│   ├── package.json
│   └── vite.config.ts
├── backend/            # Flask server
│   └── app.py          # API routes (/memories, /memorize)
└── sample_scripts/     # Standalone Python scripts for testing MemU API

Setup & Running

Prerequisites

• Node.js (v18+ recommended)
• Python 3.x

1. Backend Server

The backend runs on port 5001.

1. Navigate to the backend directory:

   cd backend

2. Set up environment variables:

   cp .env.example .env

   Edit .env and set your API_KEY (MemU) and OPENAI_API_KEY.

3. Install dependencies and run the server:

   pip install -r requirements.txt
   python app.py

2. Frontend Application

The frontend development server typically runs on port 5173.

cd frontend
npm install
npm run dev

API Interface Details

The Flask backend acts as a proxy to the MemU API, handling authentication and request formatting.

1. Retrieve Memories

• Endpoint: GET /memories
• Description: Retrieves relevant memories from MemU based on a query, then streams an AI-generated financial analysis (BUY/SELL/HOLD decision) based on those memories.
• Request Parameters (Query String):
  • user_id (string, required): The ID of the user.
  • agent_id (string, optional): The ID of the agent context.
    • Note: If not provided, the backend attempts to infer the relevant currency pair (e.g., "USD/JPY") from the query using OpenAI and formats it as agent_USD/JPY.
  • query (string, required): Search query text to find relevant memories and ask for advice.
• Response Format: Server-Sent Events (SSE)
  • Event: memories: JSON object containing retrieved memories from MemU API.
  • Event: token: Streamed text tokens from OpenAI providing the analysis (Action and Reason).
• Example Usage:

  curl -N "http://localhost:5001/memories?user_id=demo_user&query=Should%20I%20buy%20USD/JPY?"

2. Memorize Interaction

• Endpoint: POST /memorize
• Description: Stores a user interaction into MemU memory.
• Request Body (JSON):
  • user_id (string, required): The ID of the user.
    • Note: Backend automatically prepends user_ prefix before sending to MemU.
  • agent_id (string, required): The ID of the agent context (e.g., currency pair).
    • Note: Backend automatically prepends agent_ prefix before sending to MemU.
  • user_message (string, required): The content text to be memorized.
• Response: JSON object confirming the memory task submission from MemU API.
• Example Usage:

  curl -X POST http://localhost:5001/memorize \
       -H "Content-Type: application/json" \
       -d '{
         "user_id": "demo_user",
         "agent_id": "USD/JPY",
         "user_message": "The resistance level is 150.00"
       }'

Environment Configuration

The application requires the following environment variables in backend/.env:

• API_KEY: Your MemU API Key (v3).
• OPENAI_API_KEY: Your OpenAI API Key.