Xore - Advanced Crypto Intelligence System

Overview

Xore is an AI-driven cryptocurrency intelligence platform that combines Market Analysis, On-Chain Metrics, News Sentiment, and Deep Learning Models (GRU/Transformers) to provide actionable trading insights. The system features a multi-agent architecture (CrewAI) where specialized agents collaborate to analyze data and generate signals.

🚀 Key Features

• Multi-Agent System: Specialized agents for Market, On-Chain, News, and Trading logic.
• Deep Learning: Hybrid Transformer-GRU models for price prediction.
• Real-Time Data: Live integration with LiveCoinWatch, Ankr (On-Chain), and Google Search (News).
• Interactive UI: Next.js frontend with dynamic chat, real-time widgets, and educational "Learn Mode".
• Local LLM Support: Optimized for DeepSeek R1 and Llama 2 via vLLM.

---

📂 Repository Structure

The project is divided into two main components:

• frontend/: The Next.js web application.
• backend/: The core application logic, agents, and API.

Backend Documentation

The backend logic is organized into modular directories. Please verify the README.md in each subfolder for detailed file documentation:

• Agents: Autonomous agents (Market, News, OnChain, Trading).
• Models: Financial models and trading strategies.
• Services: Core business logic (LLM management, market data, XP system).
• API: FastAPI endpoints and routers.
• Utils: Shared utilities, API clients, and helpers.
• Config: Application configuration and settings.
• Tokens: Data structures for inter-agent communication.

---

🛠️ Usage

Prerequisites

• Linux Environment with GPU support (CUDA).
• Python 3.10+
• Node.js 18+
• vLLM compatible GPU.

Local LLM Setup

The system requires a local LLM model to function offline or with high privacy.

1. Download Model: Download DeepSeek-R1-Distill-Llama-8B (or your preferred compatible model) from HuggingFace.
2. Directory: Place the model files in backend/llm-models/DeepSeek-R1-Distill-Llama-8B.
3. Verify Path: Ensure start_vllm.sh points to this directory.

   # Example directory structure
   backend/llm-models/
   └── DeepSeek-R1-Distill-Llama-8B/
       ├── config.json
       ├── model.safetensors
       └── tokenizer.json

Starting the System

Run the master script to start Backend, Frontend, and LLM Server:

./start_all.sh

Accessing the App

• Web UI: http://localhost:3000
• API Docs: http://localhost:8000/docs

---

🧪 Testing & Verification

• backend/notebooks/: Jupyter notebooks for EDA and model training.

📝 Notes

• Live Data: The system uses real live data (LiveCoinWatch, Ankr).
• Asset Awareness: The system dynamically handles BTC and ETH.