Agent Settings Guide

This guide covers everything you need to configure AI agents in Wegent through the Web interface. An Agent combines Bots, collaboration modes, and AI models to accomplish tasks.

📋 Table of Contents

Overview
Agents in collaboration projects
Accessing Agent Settings
Creating an Agent
Shared Base Capabilities
Configuring Bots
Collaboration Modes
Model Configuration
MCP Tools & Skills
Best Practices
FAQ

🎯 Overview

What is an Agent?

An Agent is your AI assistant in Wegent. It can:

Execute development tasks
Answer questions
Review code
Generate documentation
And much more...

Architecture

Agent = Bot(s) + Collaboration Mode

Bot = Executor + Model + Prompt + Tools

Component	Description	Example
Agent	User-facing AI assistant	"Code Assistant"
Bot	Building block of an Agent	"Frontend Developer Bot"
Executor	Runtime environment	ClaudeCode, Dify, Chat
Model	AI brain	Claude Sonnet 4, GPT-4o
Prompt	Personality & expertise	System instructions
Tools	External capabilities	GitHub MCP, Skills

🧩 Agents in collaboration projects

Wegent Web and Wework use the same agent creation form in collaboration projects. The form keeps four independent dimensions:

Dimension	Meaning	Examples
Agent source	Where the agent definition is stored	Current project, shared agent
Hosting mode	Which class of environment hosts the agent	Device execution, Wegent-managed
Executor type	Which runtime the agent uses	Codex, Claude Code
Capability source	How Skills, Plugins, and MCP are obtained	Follow the runtime device, manually selected capabilities

These four dimensions are independent. Wegent-managed is a hosting mode, not an Agent source or executor type. A shared agent retains its hosting mode, executor, and capability configuration in its own definition, while the project stores a reference to it. A shared agent may be dispatched to an eligible device or hosted by Wegent. An agent created in the current project stores its executor, model, system prompt, and capability requirements directly from the form.

Capability sources

Follow the runtime device: the task uses the Plugins, Skills, MCP servers, and local capabilities available to the current user on whichever device runs it. The agent is not permanently bound to the device used during creation.
Manually select capabilities: the selected Plugins, Skills, and MCP configuration are saved with the agent. Before a task starts, the system selects or prepares an eligible execution device and synchronizes those capabilities into its runtime.

The system prompt editor uses the same reference interaction as the task composer. Supported capability references can be inserted with @ or /. A reference expresses a capability requirement; it does not implicitly bind the agent to the current computer.

⚙️ Accessing Agent Settings

Via Web Interface

Log in to Wegent
Click Settings in the sidebar
You'll see these tabs:

Tab	Description
Team	Manage your agents
Bot	Manage individual bots
Models	Configure AI models
Shells	Custom executors
Skills	Chat and Claude Code skills
Integrations	Git tokens

Quick Access

From the chat interface, click the gear icon next to the agent selector
Or click Manage in the agent dropdown

🚀 Creating an Agent

Step 1: Open Creation Dialog

Go to Settings → Team tab
Click New Team button

Step 2: Basic Information

Field	Required	Description
Name	Yes	Agent name (e.g., "Code Assistant")
Description	No	What this agent does
Icon	No	Visual identifier
Bind Mode	Yes	Where agent appears (Chat/Code)
Requires Repository	No	Need code repo to work?

Bind Mode Options

Mode	Description
Chat	Appears in Chat mode for conversations
Code	Appears in Code mode for development
Both	Available in both modes

Step 3: Select Collaboration Mode

Choose how your agent works:

Mode	Bots	Best For
Solo	1	Simple tasks, beginners
Pipeline	2+	Sequential workflows
Coordinate	2+	Parallel analysis

The web UI currently offers Solo, Pipeline, and Coordinate when creating or editing agents.

Recommendation: Start with Solo mode for simplicity.

Step 4: Configure Bot

For Solo mode, configure the bot directly:

Name: Bot identifier
Executor: Select runtime (ClaudeCode, Chat, Dify)
Bind Model: Select AI model
Prompt: Define personality and expertise
MCP Config: Add external tools (optional)
Skills: Add capabilities (optional)

For other modes, select existing bots or create new ones.

Step 5: Save

Click Save to create your agent. It will appear in your agent list.

🧰 Shared Base Capabilities

When an agent is created with the simple form, Use shared base capabilities is enabled by default. The custom agent reuses the general Skills, MCP tools, and plugins configured on the system default agent while keeping its own name, system prompt, model, and executor.

This works with Chat, Claude Code, Codex, and other supported executors; it does not require switching to Chat Shell.
Wegent merges the base capabilities into the custom agent configuration and performs one model execution. It does not invoke the default agent first and the custom agent second.
Same-name MCP and plugin entries from the custom agent take precedence, while Skills are deduplicated.
Only one inheritance level is supported to keep capability resolution predictable.
Enabling the option provides more capabilities but can increase context size, tool discovery, and tool-call usage. Disable it for pure chat or minimal executions that do not need these capabilities.
In Follow runtime device mode, Wegent uses the current user's capabilities on the execution device and does not load fixed Ghost capabilities.

After saving, the UI continues to show the Skills available to the agent. During execution, the merged capability configuration is cached within the request so prompt, Skill, MCP, and plugin assembly do not repeat the same reads.

🤖 Configuring Bots

Bot Components

Component	Required	Description
Name	Yes	Unique identifier
Executor	Yes	Runtime environment
Model	No	AI model (can inherit)
Prompt	No	System instructions
MCP Config	No	External tools
Skills	No	Chat and Claude Code skills

Executor Types

Executor	Description	Use Case
ClaudeCode	Claude Code SDK in Docker	Code development
Chat	Direct LLM API	Simple conversations
Dify	External Dify API	Dify workflows

Writing Effective Prompts

Structure:

You are a [role], skilled in [expertise].

Your responsibilities:
- [Task 1]
- [Task 2]

Working principles:
- [Guideline 1]
- [Guideline 2]

Example:

You are a senior frontend engineer, skilled in:
- React 18+ and TypeScript
- Tailwind CSS
- Performance optimization

Your responsibilities:
- Develop high-quality UI components
- Write clean, maintainable code
- Follow React best practices

Working principles:
- Use functional components and Hooks
- Write TypeScript types for all components
- Focus on user experience

Managing Bots

Go to Settings → Bot tab
View all your bots
Click Edit to modify, Delete to remove
Click New Bot to create standalone bots

🤝 Collaboration Modes

Solo Mode

Single bot handles everything.

User → Bot → Result

Best for: Simple tasks, getting started

Pipeline Mode

Bots execute sequentially.

Bot A → Bot B → Bot C → Result

Best for: Development → Review → Testing workflows

Configuration:

Select Leader bot (first in pipeline)
Add Member bots in order
Optionally enable Require Confirmation between stages

Coordinate Mode

Leader coordinates parallel work.

        → Bot A (parallel)
Leader → Bot B (parallel) → Aggregate
        → Bot C (parallel)

Best for: Multi-perspective analysis

🧠 Model Configuration

Binding Models to Bots

Method 1: Dropdown Selection (Recommended)

In Bot configuration, find Bind Model
Keep "Advanced Mode" OFF
Select from available models

Method 2: Advanced Mode

Toggle Advanced Mode ON
Select Model Protocol (OpenAI/Claude/Gemini)
Enter JSON configuration

Creating Custom Models

Go to Settings → Models tab
Click Create Model
Configure:
- Model Type (LLM/Embedding/Rerank)
- Protocol (OpenAI/Anthropic/Gemini)
- API Key
- Model ID

See Model Configuration Guide for details.

Per-Task Model Override

When sending a task:

Click the Model selector in chat input
Choose a different model
Enable Force Override if needed

🔧 MCP Tools & Skills

MCP (Model Context Protocol) Tools

MCP tools extend bot capabilities with external services.

Adding MCP Tools:

In Bot configuration, find MCP Config
Click Import MCP
Paste configuration JSON
Choose Replace or Append

Example - GitHub MCP:

{
  "github": {
    "command": "docker",
    "args": [
      "run",
      "-i",
      "--rm",
      "-e",
      "GITHUB_PERSONAL_ACCESS_TOKEN",
      "ghcr.io/github/github-mcp-server"
    ],
    "env": {
      "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_xxx..."
    }
  }
}

Skills

Skills add reusable capabilities to Chat and Claude Code bots.

Adding Skills:

In Bot configuration, find Skills section
Select skills from dropdown
Check the preload option next to a selected skill when it should be available from task start
Click Manage Skills to upload new ones

Preloading Skills:

Both the simple configuration UI and the full configuration UI support preload skills
Chat and ClaudeCode executors support preload skills
For ClaudeCode, preloaded skills are deployed into the Claude Code skills directory when the task starts
Preload only core skills that the bot should prioritize; leave other skills on-demand

Managing Skills:

Go to Settings → Skills tab
Upload ZIP packages containing SKILL.md
Or import from Git repositories

✨ Best Practices

1. Start Simple

Begin with Solo mode
Use one bot with clear responsibilities
Add complexity as needed

2. Choose Right Executor

Task	Executor
Code development	ClaudeCode
Simple Q&A	Chat
Dify workflows	Dify

3. Write Clear Prompts

Be specific about the role
List concrete responsibilities
Include working guidelines

4. Optimize Costs

Complexity	Model
Simple	Claude Haiku / GPT-3.5
Medium	Claude Sonnet / GPT-4o
Complex	Claude Opus / GPT-4

5. Use Descriptive Names

✅ "Frontend React Developer"
✅ "Code Review Expert"
❌ "Bot 1"
❌ "Test"

❓ FAQ

Q: What's the difference between Agent and Bot?

Agent: User-facing AI assistant (what you interact with)
Bot: Building block of an Agent

An Agent can have one Bot (Solo) or multiple Bots (other modes).

Q: Which collaboration mode should I use?

Scenario	Mode
Simple tasks	Solo
Sequential workflow	Pipeline
Parallel analysis	Coordinate

Q: Can I edit an Agent after creation?

Yes! Click Edit on any agent to modify it.

Q: Why doesn't my Agent appear in Chat/Code?

Check Bind Mode setting - enable Chat and/or Code as needed.

Click Share on the agent
Copy the share link
Others can add it to their list

Q: What's "Requires Repository"?

Enabled: Must select a code repo before starting
Disabled: Can work without a repo

Enable for development agents, disable for chat agents.

Q: Does enabling shared base capabilities make two model calls?

No. Wegent reads the custom Ghost and its base Ghost, merges their Skills, MCP tools, and plugins, and sends one final model request. The base Ghost's system prompt does not override the custom agent's identity.

Model Configuration - Set up AI models
Managing Tasks - Use agents to execute tasks
Shell Configuration - Custom executors
Skill Management - Chat and Claude Code skills

Configure your AI agents and start automating! 🚀

📋 Table of Contents​

🎯 Overview​

What is an Agent?​

Architecture​

🧩 Agents in collaboration projects​

Capability sources​

⚙️ Accessing Agent Settings​

Via Web Interface​

Quick Access​

🚀 Creating an Agent​

Step 1: Open Creation Dialog​

Step 2: Basic Information​

Bind Mode Options​

Step 3: Select Collaboration Mode​

Step 4: Configure Bot​

Step 5: Save​

🧰 Shared Base Capabilities​

🤖 Configuring Bots​

Bot Components​

Executor Types​

Writing Effective Prompts​

Managing Bots​

🤝 Collaboration Modes​

Solo Mode​

Pipeline Mode​

Coordinate Mode​

🧠 Model Configuration​

Binding Models to Bots​

Creating Custom Models​

Per-Task Model Override​

🔧 MCP Tools & Skills​

MCP (Model Context Protocol) Tools​

Skills​

✨ Best Practices​

1. Start Simple​

2. Choose Right Executor​

3. Write Clear Prompts​

4. Optimize Costs​

5. Use Descriptive Names​

❓ FAQ​

Q: What's the difference between Agent and Bot?​

Q: Which collaboration mode should I use?​

Q: Can I edit an Agent after creation?​

Q: Why doesn't my Agent appear in Chat/Code?​

Q: How do I share an Agent?​

Q: What's "Requires Repository"?​

Q: Does enabling shared base capabilities make two model calls?​

🔗 Related Resources​