Article Type: Tutorial
Audience: Workspace Administrators, Paid TeamAI Users
Last Updated: January 2026
Overview
Learn how to create custom AI agents (GPTs) tailored to your team's specific needs. This guide walks you through the complete agent creation workflow from initial setup through testing and deployment.
Learning Objectives:
Navigate the agent creation wizard from left panel navigation
Configure agent identity, visibility, and behavioral instructions
Enable built-in tools including web search, retrieval, code interpreter, and image generation
Connect knowledge sources from your datahub collections
Validate configuration through debug and preview mode
Prerequisites
You'll Need:
TeamAI paid plan with custom agent creation capabilities
Defined agent purpose and use case
Knowledge sources (collections/datastores) prepared in datahub (optional)
Step 1: Initiate Agent Creation
Access Agent Builder
Navigate to the agents tab in the left panel
Click Create agent
Result: Opens the agent creation wizard to begin configuration.
Select Creation Method
Select Create agent with AI (recommended for guided setup)
Or manually provide detailed instructions
Result: Enters the agent configuration workflow.
Step 2: Define Agent Purpose
Configure Identity and Access
Set the name for your AI agent
Write a description clarifying its role and capabilities
Select your preferred AI model
Set Visibility Level
Choose who can access your agent:
Personal - Only visible to you
Workspace - Visible to all team members in your workspace
Organization - Visible to all workspaces in your organization
Provide Behavioral Instructions
Answer: "What would you like the AI Agent to know about you to provide better responses?"
Specify how the agent should behave and respond
Define tasks to focus on
Include actions to avoid
Result: Establishes agent identity, access controls, and behavioral foundation.
Important: Choose visibility carefully based on data sensitivity and intended audience.
Step 3: Configure Conversation Features
Enable Conversation Opener
Toggle Conversation Opener to enabled (recommended)
Write an Opening Statement (example: "Hello! I'm here to help you stay informed about your student's academic journey.")
Personalize the statement to match your agent's purpose
Add Opening Questions
Create questions the AI can ask users to start conversations
Personalize questions for your use case
Use AI-generated suggestions or write your own
Result: Creates an engaging entry point that establishes connection with users.
Tip: Effective opening questions help guide users toward successful interactions.
Step 4: Enable Built-in Tools
Select Agent Capabilities
Enable tools based on your agent's requirements:
Tool | Purpose | When to Enable |
Web Search 🔍 | Access real-time information from the web | When agent needs current events, market data, or live information |
Retrieval 📄 | Extract information from specific websites | For competitor analysis, article summarization, or product research |
Code Interpreter 💻 | Automate spreadsheet tasks and calculations | For data analysis, custom calculations, or technical problem-solving |
Image Generation 🎨 | Create professional visuals with Imagen | For marketing materials, mock-ups, or social media graphics |
Toggle each tool on or off based on your needs
Review tool descriptions to understand capabilities
Result: Extends your agent's capabilities beyond basic conversation.
Tip: Only enable tools your workflow requires to maintain focus and efficiency.
Step 5: Connect Knowledge Sources
Link Datahub Collections
Navigate to the Connect datahub section
Select collections and datastores your agent can access
Use Sync tree selection to select/unselect nested datastores with parent nodes
Result: Provides your agent with organizational knowledge to reference in responses.
Privacy Note: Review sensitive information before adding datastores, as content may appear in agent outputs.
Step 6: Review and Validate Configuration
Verify Agent Settings
Review the configuration summary containing:
Name and Description
Model selection
Pre-built Tools enabled (web_search, retrieve, code_interpreter, imagen)
Custom Tools (if created)
Knowledge Sources (collections selected)
Instructions provided
Visibility settings
Result: Confirms all parameters are correct before final creation.
Important: Carefully review all settings. Changes after creation may require republishing.
Step 7: Test and Deploy
Debug and Preview
Access debug and preview mode
Test your agent with sample queries
Refine instructions and settings based on test results
Click Publish when validation is complete
Result: Your agent is deployed and available to team members based on visibility settings.
Best Practices
Start with Clear Purpose: Define your agent's goal before configuration to ensure alignment with business needs
Choose Appropriate Visibility: Consider data sensitivity when selecting access levels—Personal for private work, Workspace for team collaboration, Organization for company-wide access
Write Specific Instructions: Clear, detailed instructions produce more accurate and consistent responses
Enable Tools Strategically: Only activate tools your workflow requires to avoid unnecessary complexity and potential errors
Test with Real Scenarios: Use actual questions your team will ask to validate agent performance
Organize Knowledge Sources: Well-structured datastores improve retrieval accuracy and response quality
Review Before Publishing: Double-check all settings in the review screen to prevent configuration errors
Monitor and Iterate: Track usage patterns and refine agent based on feedback and performance metrics
Troubleshooting & FAQ
Q: My agent cannot access the datastores I selected.
A: Verify sync tree selection includes all nested datastores and confirm datahub permissions are properly configured.
Q: Team members cannot see my published agent.
A: Confirm visibility is set to "Workspace" or "Organization" (not "Personal") and validate workspace permissions.
Q: How do I change my agent's model after creation?
A: Edit the agent configuration, select a different model, then republish to apply changes.
Q: Can I disable tools after publishing?
A: Yes, modify tool settings in the configuration and republish the agent to update capabilities.
Q: My agent is not asking the opening questions I configured.
A: Ensure "Conversation Opener" is enabled and verify the opening statement is properly formatted.
Q: What is the difference between Retrieval and Web Search tools?
A: Retrieval extracts information from specific, pre-configured websites; Web Search queries the live internet broadly.
Q: How do I remove a datastore connection?
A: In the datahub connection screen, use sync tree selection to deselect the datastore.
Q: Can I preview my agent before publishing?
A: Yes, use debug and preview mode to test and refine before making the agent available to your team.