Skip to main content

Understanding Agents

"Agents" might sound technical, but the concept is simple: an agent is an AI assistant with a specific job and personality. Just like a company might have a receptionist, an accountant, and a designer — you can create different AI agents for different tasks.


What Exactly Is an Agent?

An agent is an AI that has been given:

  1. A name — So you can identify it (e.g., "Code Reviewer", "Email Writer")
  2. Instructions — What it should do and how it should behave (called a "system prompt")
  3. An AI model — Which AI brain to use (Claude, GPT-4, Llama, etc.)
  4. Skills — Special abilities like web search, file reading, or code execution
  5. Settings — How creative or precise it should be

Think of it like hiring an employee: you tell them their role, give them their tools, and set expectations.


Your Default Agent

ClawDesk comes with a default agent already set up. When you start chatting, you're talking to this default agent. It's a general-purpose assistant that can help with almost anything.

But the real power comes from creating specialized agents.


Why Create Specialized Agents?

The same AI model can behave very differently depending on the instructions you give it:

A coding agent told to "always write clean, documented code" gives much better code than a generic assistant.

A writing agent told to "use vivid imagery and vary sentence length" produces more engaging prose.

A support agent told to "be empathetic and break solutions into numbered steps" handles customer issues better.


Creating Your First Agent

Here's how to create an agent in ClawDesk:

Step 1: Go to the Agents Page

Click "Agents" in the sidebar.

Step 2: Click "New Agent"

You'll see the Agent Designer — a form where you define your agent.

Step 3: Fill in the Details

FieldWhat to EnterExample
NameA short name for your agent"Email Helper"
DescriptionWhat this agent does"Helps draft professional emails"
ModelWhich AI to useClaude claude-sonnet-4-20250514
System PromptInstructions for the agentSee examples below
TemperatureCreativity level (0-1)0.7 for creative, 0.3 for precise

Step 4: Write a Good System Prompt

The system prompt is the most important part. It's the set of instructions that defines how your agent behaves. Here are some examples:

Email Writer Agent

You are an expert email writer. When given a topic or situation, draft a 
professional, clear email. Always:
- Start with a friendly greeting
- Get to the point quickly
- Use a professional but warm tone
- End with a clear call to action
- Keep emails under 200 words unless asked otherwise

Ask clarifying questions if the request is vague.

Code Reviewer Agent

You are a senior software engineer reviewing code. For every piece of code 
you review:
1. Check for bugs and potential errors
2. Look for security vulnerabilities
3. Suggest performance improvements
4. Check code style and readability
5. Rate the code quality on a scale of 1-10

Be constructive and specific. Always explain WHY something should change,
not just what to change.

Study Buddy Agent

You are a patient, encouraging study partner. When helping with learning:
- Explain concepts using simple analogies
- Ask questions to check understanding
- Break complex topics into small steps
- Celebrate progress, no matter how small
- Use examples from everyday life
- If a student is confused, try explaining differently instead of repeating

Never give direct answers to homework — guide students to figure it out
themselves.

Step 5: Choose Skills (Optional)

Skills give your agent extra abilities:

SkillGood For
Web SearchAgents that need current information
File ReaderAgents that analyze documents
Code ExecutorCoding agents that need to test code
CalculatorAgents handling math or data
BrowserAgents that need to look things up online

Step 6: Save

Click Save and your agent is ready to use. Select it in any chat.


Temperature: Creativity vs. Precision

The temperature setting controls how creative (or random) your agent is:

Agent TypeRecommended Temperature
Code helper0.2-0.3
Email writer0.5-0.6
General assistant0.7
Creative writer0.8-0.9
Brainstorming0.9-1.0

Agent Teams: Multiple Agents Working Together

ClawDesk lets you create teams of agents that work together. This is like having a small company of AI workers.

How Agent Teams Work

Example: Blog Post Pipeline

  1. You: "Write a blog post about remote work trends in 2026"
  2. Research Agent searches for the latest statistics and trends
  3. Writer Agent takes the research and drafts the article
  4. Editor Agent polishes the writing and checks for errors
  5. You receive a well-researched, well-written blog post

How to Set Up a Team

  1. Go to the Agents page
  2. Click "Team Builder" or "Pipeline"
  3. Add agents as steps in the pipeline
  4. Define the flow: which agent goes first, second, etc.
  5. Set up conditions: "If the research agent finds fewer than 3 sources, ask for more"

Agent Flows

Agent Flows are predefined setups that connect ClawDesk to external coding AI tools:

FlowWhat It Does
Claude CodeConnects to Claude's code agent for advanced programming
CodexUses OpenAI's code-specialized model
CursorIntegrates with the Cursor code editor
Shell ProcessRuns terminal commands through an agent
A2A GatewayConnects to external agent services

These flows are selected from the Flow Selector in the chat page.


Tips for Great Agents

1. Be Specific in Your System Prompts

❌ "Be helpful"
✅ "When asked about code, always explain your approach before writing code. Include comments in the code. Test edge cases."

2. Use the Right Model for the Job

  • Complex tasks → Claude or GPT-4
  • Fast simple tasks → Llama 3.1 or Mistral (local)
  • Creative writing → Higher temperature setting

3. Start Simple, Iterate

Don't try to create the perfect agent on your first try. Start with a basic system prompt, test it, and refine based on what works.

4. Give Examples in Your System Prompt

When someone asks about your hours:
GOOD: "We're open Monday-Friday, 9 AM to 5 PM EST. Is there a specific day you're planning to visit?"
BAD: "9-5"

5. Set Boundaries

Tell your agent what it should and shouldn't do:

You are a coding assistant. Only help with programming questions. 
If asked about topics unrelated to software development, politely
redirect the conversation.

What Can Agents Do? (A Summary)

CapabilityDescription
ChatHave conversations, answer questions
Execute CodeWrite and run code in multiple languages
Search the WebFind current information online
Read FilesAnalyze documents, spreadsheets, images
Browse WebsitesNavigate and extract information from web pages
Use MemoryRemember information across conversations
Work in TeamsCollaborate with other agents on complex tasks
Connect to AppsRespond on Telegram, Discord, Slack, etc.
Run on ScheduleExecute tasks at specific times
Use ToolsInterface with databases, APIs, and services via MCP

Next Steps