Skip to main content

Creating Agents

An agent is a specialized AI assistant with its own personality, instructions, and skills. Think of it like creating a custom character for a specific job.

What's the Difference Between Chat and Agents?​

Regular ChatAgent
PersonalityGeneric AI assistantCustom-defined (e.g., "friendly Python tutor")
InstructionsNoneYou write specific rules and behaviors
SkillsBasicCan have specialized tools and knowledge
Use caseQuick questionsRepeated tasks, specific roles

Example: Instead of typing "You are a friendly Python tutor who explains concepts simply" every time, create a "Python Tutor" agent with those instructions built in.

Creating Your First Agent​

Step 1: Go to the Agents Page​

Click Agents in the left sidebar. You'll see the agent roster — a list of all your agents.

Step 2: Create a New Agent​

Click the "Create Agent" or "+" button.

Step 3: Fill in the Details​

FieldWhat to enterExample
NameA short name for your agent"Writing Coach"
DescriptionWhat this agent does"Helps improve writing with clear feedback"
System PromptInstructions that tell the AI how to behaveSee examples below
ModelWhich AI model to use (or "auto" for default)"claude-sonnet-4-20250514"

Step 4: Write a System Prompt​

The system prompt is the most important part — it's like giving someone a job description. Here are some examples:

Example: Writing Coach
You are a friendly and encouraging writing coach. When the user shares their 
writing with you:

1. Start with something positive about their work
2. Identify 2-3 specific areas for improvement
3. Suggest concrete edits with explanations
4. Keep your feedback constructive and supportive
5. If they ask for help with a specific type of writing (email, essay, story), 
   tailor your advice to that format

Use simple language. Avoid jargon. Be warm and helpful.
Example: Python Tutor
You are a patient Python programming tutor for beginners. Follow these rules:

1. Always explain concepts step by step
2. Use simple analogies from everyday life
3. Show working code examples for every concept
4. When the student makes an error, explain what went wrong without judgment
5. After explaining something, ask if they want to try a practice exercise
6. Use comments in code to explain what each line does

Never assume prior programming knowledge. If they ask about something advanced, 
relate it back to basics first.
Example: Email Drafter
You are a professional email writing assistant. When the user describes an 
email they need to write:

1. Ask who the email is to and the relationship (boss, client, friend, etc.)
2. Write a professional draft
3. Keep emails concise — under 150 words unless they ask for more
4. Include a clear subject line suggestion
5. Match the tone to the situation (formal for business, friendly for colleagues)
6. If they say "make it shorter" or "more formal," adjust accordingly
Example: Research Assistant
You are a thorough research assistant. When asked about a topic:

1. Provide a clear, well-organized summary
2. Break information into sections with headers
3. Include specific facts, numbers, and dates when relevant
4. Note any important caveats or limitations
5. Suggest related topics they might want to explore
6. If you're not sure about something, say so clearly

Format your responses for easy scanning — use bullet points, headers, and 
bold text for key information.

Step 5: Save Your Agent​

Click Save. Your agent now appears in the agent roster and is available in the Chat page.

Using Your Agent​

  1. Go to the Chat page
  2. Select your agent from the agent selector at the top
  3. Start chatting — the AI will follow your agent's instructions

Agent Settings Explained​

Temperature​

Controls how creative vs. predictable the AI is:

TemperatureBehaviorGood for
0.0 - 0.3Very predictable, consistentCode, facts, math
0.4 - 0.7BalancedMost tasks
0.8 - 1.0Creative, variedWriting, brainstorming

Default: 0.7 (balanced)

Model Selection​

You can set which AI model your agent uses:

  • "auto" or "default" — Uses whatever model you pick in the chat
  • A specific model (e.g., "claude-sonnet-4-20250514") — Always uses that model, regardless of what's selected in the chat

Adding Skills to Your Agent​

Skills give your agent special abilities beyond just chatting. Go to the Skills page to browse available skills.

Examples of skills:

  • Web Search — Let the agent search the internet
  • File Reading — Let the agent read files on your computer
  • Code Execution — Let the agent run code
  • Document Analysis — Let the agent analyze PDFs and documents

To add a skill to an agent:

  1. Go to Agents → select your agent → Edit
  2. In the skills section, check the skills you want to enable
  3. Save

Agent Teams​

For advanced use cases, you can create agent teams — multiple agents working together:

RoleWhat it doesExample
RouterDecides which agent handles each message"Send coding questions to the Coder, writing to the Writer"
ResearcherGathers information"Find background information on the topic"
WriterProduces content"Write the final article using the research"
ReviewerChecks quality"Review the article for errors"

Creating a Team​

  1. Go to Agents → click "Create Team"
  2. Add agents to the team
  3. Assign roles (router, researcher, writer, reviewer, etc.)
  4. The Team Builder Canvas lets you visually arrange how agents connect

This is an advanced feature — start with single agents first and explore teams when you're comfortable.

Managing Your Agents​

Edit an Agent​

Click on an agent in the roster → edit its name, prompt, model, or skills → save.

Delete an Agent​

Click on an agent → click Delete. This is permanent.

Duplicate an Agent​

A quick way to create a similar agent — duplicate an existing one and modify it.

Tips for Great Agents​

  1. Be specific in your system prompt — The more detailed your instructions, the better the agent performs
  2. Test and iterate — Chat with your agent, see how it responds, and refine the prompt
  3. Start simple — Begin with a short prompt and add detail as needed
  4. One job per agent — It's better to have several focused agents than one that tries to do everything
  5. Set the right temperature — Use lower for factual tasks, higher for creative ones