Technical Architecture

GitHub - 0xdevpro/deepcore: AI Agent FrameworkGitHub

DeepCore is built on the innovative MCP (Model-Controller-Protocol) architecture, a design pattern that enables us to build highly flexible and powerful intelligent agent systems.

DeepCore's architecture consists of three main layers:

1. Web3 Agent Store Layer

The top layer provides a marketplace of specialized agents for different purposes:

• Analysis Agent - For data analytics and insights generation

• Trade Agent - For executing trading strategies on various platforms

• Media Agent - For content creation and media interaction

• DeepResearch Agent - For in-depth research and knowledge discovery

• Additional specialized agents - Extensible for various domain-specific tasks

2. DeepCore Agent Protocol Layer

The core protocol layer is where the main agent intelligence and orchestration happens:

Service Components

• MCP Service - Implements the Model-Controller-Protocol pattern

• SSE (Server-Sent Events) - Provides real-time communication

• CMD - Command interface for agent control

• HTTP Service - RESTful API endpoints for integrations

• OpenAPIs - Standardized API interfaces for external connectivity

• SDKs - Software Development Kits for various programming languages

Agent Orchestration

• Planner Agent - Central coordinator that breaks down complex tasks

• TaskAgents - Specialized agents that execute specific subtasks

• Tools Integration - Various tool categories available to agents:

  • CodeAct - For code generation and execution

  • Browser - For web browsing and information retrieval

  • Initial Tools - Basic built-in tooling

  • Search - Search capabilities across various sources

  • Custom Tools - User-defined or domain-specific tools

Client Integration

• Tools Center - Central registry for tool discovery and management

• Authorization - Security and permissions management

• MCP Service for Client - Client-facing interfaces for various platforms (APP | WEB | Desktop)

3. Chain Foundation Layer

The bottom layer provides blockchain and data infrastructure:

• Multi-chain Support - Integration with major blockchains (BASE, BTC, ETH, BNB, SOL, APT, SUI, etc.)

• Social Media Integration - Connections to platforms like X and Telegram

• DeFi Integration - Support for DEX and CEX interactions

• Third-party Platform Support - Extensible integration with external platforms

Core Components

Agent System

DeepCore's agent system consists of the following main components:

• Agent Core - Core agent logic implementation, managing reasoning processes and tool invocation

• Memory System - Short-term and long-term memory management, supporting context awareness and historical queries

• Tool Manager - Tool registration, validation, and execution management

• Prompt Engine - Advanced prompt templates and prompt optimization

• LLM Connector - Multi-model interface, supporting model mixing and fallback strategies

Tool Integration

DeepCore supports multiple tool types:

• OpenAPI Tools - Automatically integrate RESTful APIs through OpenAPI specifications

• Blockchain Tools - Specialized tools for interacting with various blockchain networks

• Data Analysis Tools - Tools for processing and analyzing large amounts of data

• Custom Tools - Support for developers to create and register custom tools

Security Mechanisms

DeepCore implements multi-layered security mechanisms:

• Permission Control - Fine-grained API access permission management

• Resource Limits - Monitoring and limiting agent resource usage

• Audit Logs - Comprehensive operation logging

• Vulnerability Protection - Mechanisms to prevent common security vulnerabilities