Agent Builder - Utari Docs

Overview

The Agent Builder enables your Utari workers to create entirely new AI agents from scratch through conversation. This meta-capability allows workers to design, configure, and deploy specialized agents with custom instructions, integrated services, scheduled automation, and tailored tool sets—all without manual configuration interfaces.

User Permission Required: Agent Builder MUST ask for explicit user confirmation before creating any new agent. Workers will provide details about the planned agent and wait for your approval before proceeding.

Agent Builder Capabilities

Create Agent

Build completely new AI agents with custom configurations and capabilities

Update Config

Modify existing agents including prompts, tools, and settings

MCP Integration

Search, configure, and connect external services via MCP

Credential Management

Create and configure credential profiles for service authentication

Schedule Triggers

Set up automated execution on schedules using cron expressions

Discover Tools

Find available tools for authenticated services

Creating a New Agent

Agent Creation Process

Define Requirements

Describe what you need the agent to do:

    I need an agent that monitors competitor websites daily and generates reports
    
    Create an agent for automating social media content distribution
    
    Build me a research agent that tracks AI industry news

Worker Proposes Agent

Your worker analyzes requirements and proposes an agent configuration:

    I'll create a "Competitor Intelligence Agent" with:
    
    Name: Competitor Monitor
    Purpose: Daily competitive intelligence tracking and reporting
    
    System Prompt:
    - Monitor competitor websites for changes
    - Track pricing, features, and announcements
    - Generate daily comparison reports
    - Flag significant changes for review
    
    Tools to Enable:
    - Web Browser (for site monitoring)
    - Web Search (for additional research)
    - Document Creator (for reports)
    - Files and Folder (for organization)
    - Spreadsheet tools (for tracking)
    
    Integrations:
    - Slack (for notifications)
    - Google Sheets (for data logging)
    
    Scheduled Trigger:
    - Daily at 8 AM
    - Prompt: "Check competitor sites and generate daily report"
    
    Shall I create this agent?

Confirm Creation

Review and approve the proposed agent:

    Yes, create this agent
    
    or
    
    Create it, but change the schedule to 9 AM and add Twitter monitoring

Worker Creates Agent

After confirmation, the worker:

Creates the new agent
Configures all specified tools
Sets up system prompts
Returns agent details and next steps

Complete Setup

    Agent created successfully! ✓
    
    Agent Name: Competitor Monitor
    Agent ID: agent-comp-123
    
    Next steps:
    1. Authenticate integrations (Slack, Google Sheets)
    2. Test with a manual run
    3. Scheduled trigger will start tomorrow at 8 AM
    
    Would you like to test the agent now?

Confirming Before Creation

Critical Requirement: Workers MUST use the ‘ask’ tool to request explicit user permission before creating agents. The confirmation request includes full details of the proposed agent configuration.

Example confirmation request:

🤖 Agent Creation Proposal

I'd like to create a new agent with the following configuration:

Name: Daily News Curator
Purpose: Aggregate and summarize AI industry news

System Prompt Additions:
- Search for AI news from top 10 sources daily
- Summarize key developments in 5 bullet points
- Categorize by: Research, Business, Policy, Applications
- Flag breakthrough announcements
- Maintain consistent format across reports

Tools:
- Web Search ✓
- Web Browser ✓
- Document Creator ✓
- Files and Folder ✓

Integrations:
- Slack (for distribution)
- Notion (for archival)

Scheduled Trigger:
- Every weekday at 7 AM
- Prompt: "Generate daily AI news digest"

Do you want me to create this agent?

Agent Configuration Components

System Prompt Design

The system prompt defines the agent’s personality, behavior, and operating procedures:

Role Definition

Purpose: Define what the agent is and doesExample:

    You are a Competitive Intelligence Analyst specializing in SaaS companies. Your role is to monitor competitor activities, track market trends, and provide actionable insights to the product and marketing teams.

Behavioral Guidelines

Purpose: Establish how the agent should operateExample:

    - Always cite sources for competitive data
    - Present findings objectively without bias
    - Flag significant changes immediately
    - Maintain consistent report formatting
    - Distinguish between facts and speculation
    - Respect competitor intellectual property

Workflow Instructions

Purpose: Define step-by-step processesExample:

    Daily Monitoring Workflow:
Visit each competitor website
Capture screenshots of key pages
Compare against previous day's captures
Document any changes detected
Analyze significance of changes
Generate summary report
Update tracking spreadsheet
Send alert if major changes found

Output Standards

Purpose: Specify formatting and quality expectationsExample:

    Report Format:
    - Executive Summary (3-5 bullet points)
    - Detailed Findings (by competitor)
    - Change Impact Analysis
    - Recommended Actions
    - Supporting Screenshots
    - Data Tracking Table
    
    Save to: "Competitive Intelligence/Daily Reports/[DATE]"

Domain Expertise

Purpose: Embed specialized knowledgeExample:

    Focus Areas:
    - Pricing changes and packaging updates
    - New feature announcements
    - Marketing messaging shifts
    - Customer case studies
    - Partnership announcements
    - Team expansions (especially leadership)
    - Funding or acquisition news

Tool Selection Strategy

Research Agents
Content Agents
Automation Agents
Analysis Agents

Essential Tools:

Web Search (finding information)
Web Browser (accessing sites)
Data Providers (structured data)
Document Creator (reports)
Spreadsheet tools (data organization)

Optional Tools:

Image Vision (analyzing screenshots)
Files and Folder (organization)

Avoid: Image generation, presentations (not needed)

Integration Configuration

Searching for MCP Servers

Search Based on Needs

    Search for MCP servers for email automation
    Find integrations for project management
    What MCP servers are available for CRM?

Review Options

Worker returns matching MCP servers:

    Found MCP Servers for Email:
    
    1. Gmail
       - Send emails
       - Read inbox
       - Manage labels
       - Search messages
       Auth: OAuth
    
    2. Outlook
       - Send/receive emails
       - Calendar management
       - Contact management
       Auth: OAuth
    
    3. SendGrid
       - Bulk email sending
       - Template management
       - Analytics
       Auth: API Key

Get Details

    Get details for Gmail MCP server
    
    Gmail MCP Server:
    Toolkit: Google (Gmail, Drive, Calendar, Sheets)
    
    Available Tools:
    - send_email: Send email messages
    - read_emails: Fetch inbox messages
    - search_emails: Search by criteria
    - create_draft: Create email draft
    - manage_labels: Organize with labels
    - get_attachments: Download attachments
    
    Authentication: OAuth 2.0
    Required Scopes: gmail.send, gmail.readonly, gmail.modify
    
    Setup Process:
    1. Create credential profile
    2. Complete OAuth flow
    3. Configure for agent
    4. Select tools to enable

Creating Credential Profiles

Create Profile for Service

    Create a credential profile for Gmail for the new agent

Receive Auth Link

    Credential profile created: gmail-daily-reporter
    
    Complete authentication:
    https://auth.composio.dev/connect/gmail-daily-reporter
    
    Click the link to:
    1. Select your Google account
    2. Grant Gmail permissions
    3. Complete OAuth flow
    4. Return here when done

Complete Authentication

Click link → Authorize → Grant permissions → Confirm

Discover Available Tools

    Discover tools for gmail-daily-reporter profile
    
    Available Gmail Tools:
    - Send email ✓
    - Read emails ✓
    - Search emails ✓
    - Create draft ✓
    - Manage labels ✓
    - Delete emails ✗ (not recommended for automation)

Configure for Agent

    Configure gmail-daily-reporter profile for Daily Reporter Agent
    
    Enable tools:
    ✓ Send email
    ✓ Read emails
    ✓ Search emails
    ✓ Create draft
    
    Configuration complete!
    Agent can now access Gmail with selected tools.

Scheduled Trigger Configuration

Creating Scheduled Triggers

Define Schedule

    Create a scheduled trigger for Daily Reporter Agent:
    - Run every weekday at 8 AM
    - Execute daily reporting workflow

Specify Execution Prompt

    Trigger Prompt:
    "Generate today's daily digest report by:
Searching for news in {{industry}}
Analyzing top 10 most relevant articles
Creating summary document
Saving to Daily Reports/{{date}}
Sending via email to {{distribution_list}}"

Set Cron Expression

    Schedule: 0 8 * * 1-5
    
    Translation: At 8:00 AM, Monday through Friday
    
    Timezone: America/Los_Angeles

Worker Creates Trigger

    Scheduled trigger created! ✓
    
    Trigger ID: trigger-daily-8am
    Agent: Daily Reporter Agent
    Schedule: Weekdays at 8:00 AM PST
    Status: Active
    
    Next execution: Tomorrow at 8:00 AM
    
    The agent will now run automatically on this schedule.

Managing Triggers

Show me all scheduled triggers for the Marketing Agent

List scheduled automation for Sales Agent

Disable the weekend reporting trigger

Pause all triggers for the Content Agent during vacation

Re-enable the daily monitoring trigger

Delete the old quarterly report trigger

Remove the outdated monitoring schedule

Complete Agent Creation Examples

Example 1: Competitive Intelligence Agent

User Request

    Create an agent that monitors our top 3 competitors daily and generates analysis reports

Worker Proposes

    I'll create a "Competitive Intelligence Agent":
    
    Name: Competitor Tracker
    
    System Prompt:
    - Monitor {{competitor_1}}, {{competitor_2}}, {{competitor_3}} daily
    - Track: pricing, features, marketing, news
    - Generate comparative analysis
    - Flag significant changes
    - Maintain historical tracking
    
    Tools:
    - Web Browser (site monitoring)
    - Web Search (news tracking)
    - Image Vision (screenshot analysis)
    - Spreadsheet tools (data tracking)
    - Document Creator (reports)
    
    Integrations:
    - Google Sheets (tracking database)
    - Slack (alerts)
    
    Scheduled Trigger:
    - Daily at 9 AM
    - "Check all competitor sites and generate daily intelligence report"
    
    Create this agent?

User Confirms

    Yes, create it

Worker Builds

    Creating Competitor Tracker agent...
    ✓ Agent created
    ✓ Tools configured
    ✓ System prompt set
    
    Next: Set up integrations
    
    1. Creating Google Sheets credential profile...
       Auth link: [URL]
       
    2. Creating Slack credential profile...
       Auth link: [URL]
       
    Please authenticate both services.

Complete Setup

    After authentication:
    
    ✓ Google Sheets connected
    ✓ Slack connected
    ✓ Tools configured for agent
    ✓ Scheduled trigger created
    
    Competitor Tracker Agent is ready!
    
    Would you like to run a test?

Example 2: Content Distribution Agent

Define Requirements

    I need an agent that takes blog posts from Notion and distributes them to Twitter, LinkedIn, and our newsletter

Agent Proposal

    I'll create a "Content Distribution Agent":
    
    Name: Content Distributor
    
    Capabilities:
    - Read new posts from Notion
    - Generate platform-specific variants:
      * Twitter: Thread format (280 char)
      * LinkedIn: Professional post
      * Newsletter: Email format
    - Post to each platform
    - Track engagement
    
    Tools:
    - Document Creator (content adaptation)
    - Image Editor (social graphics)
    
    Integrations:
    - Notion (source)
    - Twitter (distribution)
    - LinkedIn (distribution)
    - Mailchimp (newsletter)
    - Google Analytics (tracking)
    
    Trigger Options:
    - Event: When new post published in Notion
    or
    - Schedule: Check for new posts daily at 10 AM
    
    Which trigger type do you prefer?

User Selects

    Use event trigger - distribute immediately when post is published

Agent Created

    Content Distributor Agent created! ✓
    
    Configuration:
    ✓ Agent created with content adaptation capabilities
    ✓ Event trigger: Notion → New page published
    
    Required authentications:
    1. Notion: [Auth URL]
    2. Twitter: [Auth URL]
    3. LinkedIn: [Auth URL]
    4. Mailchimp: [Auth URL]
    5. Google Analytics: [Auth URL]
    
    After authentication, agent will automatically:
    - Detect new Notion publications
    - Create platform-specific content
    - Distribute across channels
    - Track engagement

Example 3: Sales Intelligence Agent

Describe Need

    Build me an agent that enriches new Salesforce leads with LinkedIn data and sends personalized outreach emails

Proposed Configuration

    I'll create a "Sales Intelligence Agent":
    
    Name: Lead Enrichment Assistant
    
    Workflow:
    1. Monitor Salesforce for new leads
    2. Research lead on LinkedIn:
       - Company info
       - Role and background
       - Recent activity
       - Mutual connections
    3. Enrich Salesforce record with data
    4. Generate personalized email using:
       - Company context
       - Pain points for their role
       - Relevant case studies
    5. Save draft in Gmail for review
    6. Log activity in tracking sheet
    
    Integrations:
    - Salesforce (lead source)
    - LinkedIn (research)
    - Gmail (outreach)
    - Google Sheets (tracking)
    
    Trigger:
    - Event-based: New lead created in Salesforce
    - Automatic execution within 5 minutes
    
    Create this agent?

Updating Existing Agents

Update Agent Configuration

Request Update

    Update my Research Agent to:
    - Add the Data Providers tool
    - Include LinkedIn in the system prompt
    - Change the name to "Market Research Agent"

Worker Applies Updates

    Updating Research Agent configuration...
    
    Changes:
    ✓ Added tool: Data Providers
    ✓ Updated system prompt: Added LinkedIn research guidelines
    ✓ Changed name: "Market Research Agent"
    ✓ Created new version: v2
    
    Previous configuration preserved in v1 for rollback if needed
    
    Market Research Agent updated successfully!

Version Control: Updates create new agent versions, preserving history. You can rollback to previous versions if needed.

Best Practices for Agent Creation

Single Responsibility

Design agents for one primary purpose rather than trying to do everything

Clear System Prompts

Write detailed, specific instructions about expected behavior and outputs

Minimal Tools

Only enable tools directly needed for the agent’s purpose

Test Before Scheduling

Run agents manually several times before setting up automated triggers

Use Template Variables

Make agents reusable with syntax in prompts and triggers

Document Purpose

Keep notes on why the agent exists and what workflows it supports

Start Simple

Begin with basic configuration, add complexity as you understand behavior

Monitor Performance

Review agent outputs regularly to ensure quality and accuracy

Troubleshooting

Worker won't create agent without confirmation

This is expected behavior:

Agent Builder REQUIRES explicit user permission
Worker will always ask for confirmation first
Review the proposed configuration
Reply with clear approval or modifications
This safety measure prevents accidental agent creation

Integration authentication fails

Try:

Opening auth link in different browser
Checking popup blockers
Verifying account permissions
Ensuring correct account selected
Requesting new credential profile if link expired

Scheduled trigger not executing

Verify:

Trigger is enabled (not disabled)
Cron expression is correct
Timezone is set appropriately
Agent has all required tools
Check trigger execution logs

Agent doesn't behave as expected

Review:

System prompt clarity and specificity
Whether all required tools are enabled
If integrations are properly authenticated
Execution logs for errors
Consider updating configuration

Can't find created agent

Check:

Navigate to Workers/Agents section
Look for the agent name
Search by agent purpose or ID
Verify creation completed successfully
Check if logged into correct workspace

Summary

You’ve successfully learned how to:

Create completely new AI agents with custom configurations

Design effective system prompts and behavioral guidelines

Select and configure appropriate tools for agent purposes

Search for and integrate MCP servers for external services

Create and configure credential profiles for authentication

Set up scheduled triggers for automated execution

Update existing agent configurations with version control

Apply best practices for reliable, effective agent design

The Agent Builder enables sophisticated meta-automation where workers can create specialized agents tailored to exact requirements. This allows you to rapidly build teams of focused agents, each optimized for specific tasks, creating a comprehensive automation ecosystem through conversational commands.

Next Steps

Agent Configuration

Learn advanced configuration techniques

MCP Server Search

Discover available integrations

Triggers & Automation

Master scheduled and event-based triggers

Credentials Manager

Manage authentication for services

​Overview

​Agent Builder Capabilities

Create Agent

Update Config

MCP Integration

Credential Management

Schedule Triggers

Discover Tools

​Creating a New Agent

​Agent Creation Process

​Confirming Before Creation

​Agent Configuration Components

​System Prompt Design

​Tool Selection Strategy

​Integration Configuration

​Searching for MCP Servers

​Creating Credential Profiles

​Scheduled Trigger Configuration

​Creating Scheduled Triggers

​Managing Triggers

​Complete Agent Creation Examples

​Example 1: Competitive Intelligence Agent

​Example 2: Content Distribution Agent

​Example 3: Sales Intelligence Agent

​Updating Existing Agents

​Update Agent Configuration

​Best Practices for Agent Creation

Single Responsibility

Clear System Prompts

Minimal Tools

Test Before Scheduling

Use Template Variables

Document Purpose

Start Simple

Monitor Performance

​Troubleshooting

​Summary

​Next Steps

Agent Configuration

MCP Server Search

Triggers & Automation

Credentials Manager

Overview

Agent Builder Capabilities

Creating a New Agent

Agent Creation Process

Confirming Before Creation

Agent Configuration Components

System Prompt Design

Tool Selection Strategy

Integration Configuration

Searching for MCP Servers

Creating Credential Profiles

Scheduled Trigger Configuration

Creating Scheduled Triggers

Managing Triggers

Complete Agent Creation Examples

Example 1: Competitive Intelligence Agent

Example 2: Content Distribution Agent

Example 3: Sales Intelligence Agent

Updating Existing Agents

Update Agent Configuration

Best Practices for Agent Creation

Troubleshooting

Summary

Next Steps