Skip to content

Latest commit

 

History

History
262 lines (194 loc) · 8.52 KB

core-concepts.md

File metadata and controls

262 lines (194 loc) · 8.52 KB
sidebar_position
1

Core Concepts

This document introduces the core concepts of the Wegent platform to help you understand each component and their relationships.

🚀 Platform Features Overview

Wegent is an open-source AI-native operating system that provides five core feature modules:

💬 Chat - AI Conversation

A fully open-source chat agent supporting:

  • Multi-model compatibility: DeepSeek, GLM, GPT, Claude, and other protocol-compatible models
  • Conversation history: Create new conversations, multi-turn dialogues, save and share history
  • Group chat: AI group conversations where AI responds based on chat history
  • Attachment parsing: Send txt, pdf, ppt, doc, and image attachments in single/group chats
  • Follow-up mode: Model helps clarify your thoughts through heuristic questions
  • Correction mode: Automatically invoke multiple models to correct answers
  • Long-term memory: Support mem0 integration for conversation memory
  • Sandbox execution: Execute commands or modify files via sandbox (E2B protocol compatible)
  • Customization: Configure prompts, MCP servers, and Skills (includes chart drawing skill)

💻 Code - Cloud Coding Engine

A cloud-based Claude Code execution engine supporting:

  • Model configuration: Configure various Claude protocol-compatible models
  • Parallel execution: Execute multiple coding tasks simultaneously in the cloud
  • Requirements clarification: AI analyzes code and questions to generate specification documents
  • Git integration: Integrate with GitHub/GitLab/Gitea/Gerrit for cloning, modifying, and creating PRs
  • MCP/Skill integration: Configure MCP servers and Skills for agents
  • Multi-turn conversations: Continue conversations with follow-up questions

📡 Follow - AI Task Trigger

A cloud-based AI task trigger supporting:

  • Full capabilities: Tasks can use all Chat and Code features
  • Scheduled/Event triggers: Execute AI tasks on schedule or based on events (e.g., summarize AI news daily at 9 AM)
  • Information feed: Display AI-generated information streams
  • Event filtering: Filter events (e.g., "only notify me if it will rain tomorrow")

📚 Knowledge - AI Document Repository

A cloud-based AI document repository supporting:

  • Document management: Upload and manage txt/doc/ppt/xls and other format documents
  • Import sources: Import web pages and DingTalk multi-dimensional tables
  • NotebookLM mode: Select documents directly in notebooks for Q&A
  • Online editing: Edit text files online in notebook mode
  • Knowledge reference: Let AI reference knowledge base in single/group chats

⚙️ Customization - Fully Configurable

All features above are customizable:

  • Web-based agent creation: Create custom agents with prompts, MCP, Skills, and multi-agent collaboration
  • Agent wizard: Automatically generate and fine-tune prompts based on requirements
  • Group sharing: Create and join groups to share agents, models, Skills, and more

🤖 Understanding Agents and Bots

What is an Agent?

An Agent is the AI assistant you interact with directly. When you create a task or start a conversation, you're working with an Agent. Think of it as your personal AI team that can help you with various tasks.

What is a Bot?

A Bot is a building block that makes up an Agent. Each Bot is configured with:

  • Persona (Ghost): Defines the Bot's personality, expertise, and available tools
  • Executor (Shell): The engine that executes tasks (Chat, Code, etc.)
  • Model: The AI model powering the Bot (GPT, Claude, etc.)

How They Work Together

Bot = Persona + Executor + Model
Agent = One or more Bots working together
Task = Agent + Your request

Example:

  • A simple chat Agent might have just one Bot
  • A development Agent might have multiple Bots: one for coding, one for code review, one for testing

🔄 Concept Relationship Diagram

graph LR
    subgraph Main["Task Execution Flow"]
        direction TB
        User["👤 You"]
        Web["🌐 Web"]
        API["🔌 API"]
        IM["💬 IM"]
        Task["🎯 Task"]
        Wegent["🌐 Wegent System"]
        
        User --> Web
        User --> API
        User --> IM
        Web --> Task
        API --> Task
        IM --> Task
        Task --> Wegent
        
        subgraph Workspace["💼 Workspace"]
            subgraph Cloud["☁️ Cloud Host"]
                ClaudeCode1["🐚 Claude Code"]
                WegentChat1["💬 Wegent Chat"]
            end
            
            subgraph PC["💻 Personal Computer"]
                ClaudeCode2["🐚 Claude Code"]
                WegentChat2["💬 Wegent Chat"]
            end
        end
        
        Wegent --> Cloud
        Wegent --> PC
    end
    
    subgraph AgentDetail["📋 Agent Structure"]
        direction TB
        AgentNote["🤖 Agent"]
        Bot1["🔧 Bot 1"]
        Bot2["🔧 Bot 2"]
        Ghost1["👻 Persona"]
        Model1["🧠 Model"]
        Ghost2["👻 Persona"]
        Model2["🧠 Model"]
        Prompt1["📝 Prompt"]
        Skill1["🎯 Skill 1"]
        Skill2["🎯 Skill 2"]
        
        AgentNote --> Bot1
        AgentNote --> Bot2
        Bot1 --> Ghost1
        Bot1 --> Model1
        Bot2 --> Ghost2
        Bot2 --> Model2
        Ghost1 --> Prompt1
        Ghost1 -.-> Skill1
        Ghost1 -.-> Skill2
        Ghost2 -.-> Skill1
    end
Loading

Note:

  • Users can submit tasks via Web, API, or IM.
  • Tasks are submitted to the Wegent system, which dispatches them to Workspaces (Cloud Host or Personal Computer).
  • Workspaces run Claude Code or Wegent Chat executors.
  • Agents consist of multiple Bots, each containing a Persona and Model. Skills (dotted lines) are loaded on-demand.

🎯 Key Components Explained

👻 Persona (Ghost)

The Persona defines what your AI assistant knows and how it behaves. It includes:

  • System instructions (e.g., "You are a helpful coding assistant")
  • Available tools and MCP servers
  • Skills that can be loaded on-demand
  • Behavioral guidelines

🐚 Executor (Shell)

The Executor determines how and where your AI executes tasks. It includes:

Executor Best For
Chat Quick conversations, Q&A
Code Programming tasks, code generation
Agno Multi-agent collaboration
Dify Workflow automation

🧠 Model

The Model is the AI brain powering your assistant:

  • Configure API keys and endpoints
  • Choose from various providers (OpenAI, Anthropic, etc.)
  • Adjust model parameters

💼 Workspace

For coding tasks, a Workspace connects your Agent to a code repository:

  • Clone from GitHub/GitLab/Gitea/Gerrit
  • Make changes and create pull requests
  • Work on specific branches

🎯 Skill

Skills are special capabilities that can be added to your AI assistant on-demand. Instead of loading all instructions at once, Skills are loaded only when needed.

Why use Skills?

  • Efficiency: Only load detailed instructions when needed
  • Modularity: Package related capabilities together
  • Extensibility: Add new abilities without changing the core agent

Examples of Skills:

  • Chart Drawing: Generate diagrams and charts using Mermaid.js
  • Code Analysis: Specialized code review capabilities
  • Data Processing: Handle specific data formats

How Skills Work:

  1. You configure which Skills are available to your Agent
  2. During a conversation, the AI decides when a Skill is needed
  3. The Skill is loaded on-demand, providing specialized instructions and tools

📖 For detailed information about Skills, see Skill System

🤝 Collaboration Modes

When an Agent has multiple Bots, they can work together in different ways:

Pipeline

Bots work in sequence, each passing results to the next.

Developer → Reviewer → Tester

Route

A leader Bot assigns tasks to the most suitable Bot.

Leader → {Frontend Bot | Backend Bot | Database Bot}

Coordinate

A leader Bot coordinates parallel work and combines results.

Leader → [Analyst, Data, Report] → Leader (combine)

Collaborate

All Bots share context and discuss freely.

[Bot A ↔ Bot B ↔ Bot C]

🔗 Related Resources

Understanding these core concepts is fundamental to using Wegent! 🚀