AI Components

AI components bring large language model (LLM) capabilities into your automations. Use them to generate content, analyze data, make decisions, and perform complex reasoning tasks.

Overview

Agent Studio provides these AI components:

Component	Purpose	Best For
LLM V2	Direct text generation	Simple prompts, content generation
LLM with Structured Output V2	Generate structured JSON	Data extraction, form filling
Pre-Built Agent	Run an agent you’ve already built	Autonomous, multi-step tasks with reusable agents
Agent V2 (legacy)	Inline autonomous reasoning with tools	Superseded by Pre-Built Agent

LLM V2

A straightforward component for generating text using a large language model.

Configuration

Field	Description
System Prompt	Instructions that define the LLM’s behavior
User Prompt	The specific request (supports @ notation)
Temperature	Creativity level (0 = deterministic, 1 = creative)

Example: Generate Email Content

System Prompt:

You are a customer success manager writing professional, friendly emails.
Keep emails concise and action-oriented.

User Prompt:

Write a brief email to @contact_name at @account_data.name.
Context: Their health score dropped from @previous_score to @current_score.
Goal: Schedule a check-in call to understand their challenges.

Output

{
  "text": "Subject: Quick Check-in - How Can We Help?\n\nHi Sarah,\n\nI noticed some recent changes in your team's usage patterns and wanted to reach out personally. I'd love to schedule a brief call this week to hear how things are going and see if there's anything we can do to better support your team.\n\nWould you have 15 minutes available Thursday or Friday?\n\nBest,\n[Your name]"
}

Use Cases

Drafting personalized emails
Generating meeting summaries
Creating task descriptions
Writing notification messages

LLM with Structured Output V2

Generates responses that conform to a specific JSON schema, ensuring consistent, parseable output.

Configuration

Field	Description
System Prompt	Instructions for the LLM
User Prompt	The request with context
Output Schema	JSON schema defining the expected output structure
Temperature	Creativity level

Example: Extract Action Items

System Prompt:

You are an assistant that extracts action items from meeting notes.

User Prompt:

Extract action items from these meeting notes:
@meeting_notes

Output Schema:

{
  "type": "object",
  "properties": {
    "action_items": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "task": { "type": "string" },
          "owner": { "type": "string" },
          "due_date": { "type": "string" },
          "priority": { "enum": ["high", "medium", "low"] }
        }
      }
    },
    "summary": { "type": "string" }
  }
}

Output

{
  "action_items": [
    {
      "task": "Send updated proposal",
      "owner": "Jane",
      "due_date": "2024-02-25",
      "priority": "high"
    },
    {
      "task": "Schedule technical review",
      "owner": "John",
      "due_date": "2024-02-28",
      "priority": "medium"
    }
  ],
  "summary": "Discussed proposal revisions and agreed on next steps for technical evaluation."
}

Use Cases

Extracting structured data from unstructured text
Classifying content into categories
Parsing meeting transcripts
Generating form-ready data

Pre-Built Agent

Drops an agent you’ve already built (in the Agents section of Agent Studio) into a flow. Instead of configuring a system prompt and tools inline, you pick a ready-made agent and give it a task — it brings its own instructions, tools, and knowledge.

Configuration

Field	Description
Agent	The pre-built agent to run (select from your agents)
Query	The task or question to send the agent
Enable Citations	Include source citations in the agent’s response

If the selected agent uses tools that normally require approval, adding it prompts “Auto-approve these tools?” — because a flow runs unattended and can’t pause for in-loop approval, confirming lets those tools run without prompting. See Third-Party MCP Servers.

See Agents for building the agents you select here.

Agent V2

Legacy. Agent V2 has been superseded by the Pre-Built Agent component above. Existing flows keep working, but build new agent steps with Pre-Built Agent.

An autonomous agent that can reason, use tools, and perform multi-step tasks — you configure its instructions and tools inline on the component.

How Agents Work

Unlike simple LLM calls, agents can:

Reason about what steps to take
Use tools to fetch information or take actions
Iterate through multiple steps to complete a task
Adapt based on intermediate results

Agent receives task
    ↓
Agent plans approach
    ↓
Agent calls tools as needed
    ↓
Agent synthesizes results
    ↓
Agent returns final answer

Configuration

Field	Description
System Prompt	Define the agent’s role and behavior
User Prompt	The task to accomplish
Tools	Select which tools the agent can use
Max Iterations	Limit on reasoning steps

Available Tools

Agents can be configured with various tools:

Tool Category	Examples
Data Access	Query accounts, fetch tasks, get activities
Communication	Send emails, post to Slack
CRM	Create Salesforce cases, update opportunities
Analysis	Calculate metrics, generate insights

Example: Research and Summarize Account

System Prompt:

You are a customer success analyst. Research accounts thoroughly
and provide actionable summaries. Use available tools to gather
information before making recommendations.

User Prompt:

Research @account_data.name and prepare a summary for the upcoming QBR.
Include:
- Recent activity trends
- Health indicators
- Key risks and opportunities
- Recommended talking points

Tools Enabled:

Get account activities
Get account health history
Get open tasks
Get contact details

Output

The agent will:

Fetch recent activities for the account
Analyze health score trends
Review open tasks and their status
Identify key contacts
Synthesize into a structured QBR summary

Use Cases

Complex research tasks
Multi-step workflows
Decision-making with multiple data sources
Autonomous customer analysis

Choosing the Right AI Component

Use LLM V2 when...

You need simple text generation
The task is straightforward
You want fast, single-step output
Format flexibility is acceptable

Use Structured Output when...

You need specific data formats
Output will be used by other nodes
You need reliable parsing
Consistency is important

Use a Pre-Built Agent when...

Task requires multiple steps
You need to fetch/combine data
Complex reasoning is required
You want a reusable agent with its own tools

Best Practices

Prompt Engineering

Be specific and clear

Good prompt:

Analyze the health score trend for this account over the past 90 days.
If the score dropped more than 10 points, identify potential causes
based on recent activities. Format as bullet points.

Poor prompt:

Tell me about this account's health.

Provide context

Always include relevant context in prompts:

Account information
Recent activities
Historical data
Your specific goals

Use examples

For complex outputs, provide examples:

Generate a risk assessment in this format:
Risk Level: [High/Medium/Low]
Key Factors:
- Factor 1
- Factor 2
Recommended Actions:
- Action 1
- Action 2

Temperature Settings

Temperature	Behavior	Use For
0.0 - 0.3	Deterministic, focused	Data extraction, analysis
0.4 - 0.6	Balanced	General tasks
0.7 - 1.0	Creative, varied	Content generation, brainstorming

Error Handling

Validate AI outputs

Add condition nodes after AI components to validate:

Output is not empty
Required fields are present
Values are within expected ranges

Have fallback paths

Create alternative flows for when AI doesn’t produce usable output:

AI Component → Condition (valid output?) → True: Continue
                                         → False: Use default/alert

Troubleshooting

Output is too generic

Add more specific context to your prompt
Include examples of desired output
Lower the temperature
Provide relevant data via @ notation

Output format is wrong

Use Structured Output component instead of plain LLM
Add explicit format instructions to the prompt
Include an example in the prompt

Agent takes too long

Reduce the number of enabled tools
Lower max iterations
Simplify the task into smaller steps
Add more guidance to narrow the search space

Agent doesn't use tools

Verify tools are enabled in configuration
Add explicit instructions to use specific tools
Check that the task actually requires tools

Next Steps

Agents Overview

Build standalone AI agents

Testing Automations

Test your AI components

Getting Started

Automations

AI Agents

AI Components

AI Components

Overview

LLM V2

Configuration

Example: Generate Email Content

Output

Use Cases

LLM with Structured Output V2

Configuration

Example: Extract Action Items

Output

Use Cases

Pre-Built Agent

Configuration

Agent V2

How Agents Work

Configuration

Available Tools

Example: Research and Summarize Account

Output

Use Cases

Choosing the Right AI Component

Use LLM V2 when...

Use Structured Output when...

Use a Pre-Built Agent when...

Best Practices

Prompt Engineering

Temperature Settings

Error Handling

Troubleshooting

Next Steps

Agents Overview

Testing Automations

​AI Components

​Overview

​LLM V2

​Configuration

​Example: Generate Email Content

​Output

​Use Cases

​LLM with Structured Output V2

​Configuration

​Example: Extract Action Items

​Output

​Use Cases

​Pre-Built Agent

​Configuration

​Agent V2

​How Agents Work

​Configuration

​Available Tools

​Example: Research and Summarize Account

​Output

​Use Cases

​Choosing the Right AI Component

Use LLM V2 when...

Use Structured Output when...

Use a Pre-Built Agent when...

​Best Practices

​Prompt Engineering

​Temperature Settings

​Error Handling

​Troubleshooting

​Next Steps

Agents Overview

Testing Automations

AI Components

Overview

LLM V2

Configuration

Example: Generate Email Content

Output

Use Cases

LLM with Structured Output V2

Configuration

Example: Extract Action Items

Output

Use Cases

Pre-Built Agent

Configuration

Agent V2

How Agents Work

Configuration

Available Tools

Example: Research and Summarize Account

Output

Use Cases

Choosing the Right AI Component

Best Practices

Prompt Engineering

Temperature Settings

Error Handling

Troubleshooting

Next Steps