Agents Configuring

Agent Configuration

This guide covers configuration options for agents in GT AI OS.

Editing Agent Configuration

Accessing Edit Mode

1. Go to the agent's detail page
2. Click Edit
3. Make your changes
4. Click Save

Agent Icon (Added in 2.0.36)

Customize your agent's visual identity with a custom icon that appears on agent cards and in chat conversations.

Uploading a Custom Icon

1. In edit mode, find the Agent Icon section
2. Click Upload Image
3. Select a PNG or JPEG image (max 2MB)
4. The preview updates immediately
5. Save your changes

Icon Requirements

Requirement	Value
File formats	PNG, JPEG
Maximum size	2MB
Recommended dimensions	Square (e.g., 256x256, 512x512)

Removing an Icon

1. In edit mode, find the Agent Icon section
2. Click Remove next to the current icon
3. The agent reverts to the default shape
4. Save your changes

Tips for Good Icons

• Use a square image for best results
• Keep the design simple and recognizable at small sizes
• Ensure good contrast for visibility
• Consider using your organization's branding

Disclaimer

Add an optional disclaimer message that displays in the chat interface when users interact with this agent. This is useful for setting expectations and providing important context.

Configuring a Disclaimer

1. In edit mode, find the Disclaimer section
2. Enter your message in the text field (max 500 characters)
3. Save your changes

What to Include

• Usage guidelines and limitations
• Important context about the agent's purpose
• Any legal or compliance notices
• Expectations about response accuracy

Example Disclaimers

• "This agent is designed for general information purposes. Always verify critical information with official sources."
• "Responses are generated by AI and may not always be accurate. Please review outputs before taking action."

LLM Configuration

Select which Large Language Model (LLM) powers your agent. Different models have varying capabilities, costs, and performance characteristics.

Selecting a Model

1. In edit mode, find the LLM Configuration section
2. Click the model dropdown
3. Select your desired model
4. Review the token limits and pricing displayed
5. Save your changes

Model Information

When you select a model, you'll see:

Information	Description
Provider	The model provider (e.g., OpenAI, Groq, Ollama)
Max Context	Total token limit for input + output
Max Output Tokens	Maximum tokens the model can generate
PAYG Pricing	Cost per million tokens (input/output)

Choosing the Right Model

• For accuracy: Choose larger models with more parameters
• For speed: Choose smaller, faster models
• For cost: Consider pricing and your expected usage
• For privacy: Consider local models (Ollama) for sensitive data

Model Parameters

Fine-tune how your agent generates responses.

Temperature

Controls how predictable vs. creative the agent's responses are:

Range	Behavior	Best For
0.0-0.3	Focused, consistent, deterministic	Factual Q&A, technical support, data analysis
0.4-0.7	Balanced	General use (0.7 is the default)
0.8-2.0	Creative, varied, surprising	Brainstorming, creative writing, ideation

Tip: Start with 0.7 and adjust based on results. Lower if you want more consistent answers; raise if responses feel too repetitive.

System Prompt Configuration

The system prompt defines your agent's personality, capabilities, and behavior. This is the most important configuration for shaping how your agent interacts with users.

Writing Effective Prompts

Your system prompt should include:

1. Role Definition: Who the agent is
2. Capabilities: What it can help with
3. Constraints: What it shouldn't do
4. Tone: How it should communicate
5. Format: How to structure responses

Example Structure

# Role
You are [role description].

# Capabilities
You can help with:
- [capability 1]
- [capability 2]

# Guidelines
- [guideline 1]
- [guideline 2]

# Response Format
[formatting instructions]

Best Practices

• Be specific about the agent's role and expertise
• Define clear boundaries for what the agent should not do
• Specify preferred response formats (bullet points, paragraphs, etc.)
• Include examples of ideal responses when helpful

Dataset Attachments

Connect your agent to datasets for retrieval-augmented generation (RAG), allowing it to reference your organization's knowledge.

Managing Attached Datasets

1. In edit mode, scroll to the Dataset Selection section
2. Use the search box to find datasets
3. Check boxes to attach, uncheck to detach
4. Click the X on dataset badges to remove them
5. Save changes

Dataset Selection Tips

• Only attach relevant datasets
• Consider dataset size and quality
• Update attachments when datasets change
• The agent can search attached datasets for relevant information during conversations

Notes on Model Compatibility

Some models (like compound models) may not support RAG functionality. When this is the case, the dataset selection will be disabled with an explanation.

Easy Buttons

Easy buttons provide quick-access conversation starters (up to 10):

Configuring Easy Buttons

1. In edit mode, find Easy Buttons
2. Click Add Easy Button to create new buttons
3. These appear as buttons in chat
4. Click the X to remove individual buttons

Example Easy Buttons

• "Summarize this document"
• "Explain this code"
• "Help me draft an email"

Export and Import

For exporting and importing agents, see Sharing Agents.

Troubleshooting

Configuration Not Saving

• Verify all required fields are filled
• Check for validation errors
• Ensure you have edit permissions

Agent Behaving Unexpectedly

• Review system prompt for clarity
• Check model parameters
• Verify attached datasets
• Test with simple queries first