AI Capabilities

Overview of AI features in Agents

Agents provides powerful AI capabilities that enable intelligent automation. This section covers how to leverage AI effectively in your workflows.

AI in Workflows

AI transforms static automation into intelligent workflows that can:

Analyze meeting transcripts and documents
Summarize conversations and key points
Extract action items, decisions, and insights
Generate personalized content and messages
Classify information and route accordingly
Decide next steps based on context

AI Node Types

AI Prompt

For structured AI processing with defined outputs:

text

[AI Prompt]
├── Input: Messages array with context
├── Output: Structured data (string, integer, boolean, etc.)
└── Use for: Summarization, extraction, classification

Best for:

Generating summaries
Extracting specific information
Classification tasks
Simple transformations

AI Agent

For complex tasks requiring multi-step reasoning:

text

[AI Agent]
├── Input: Messages array with instructions
├── Tools: List, Search, Open, Web Search
├── Output: Structured data with reasoning
└── Use for: Research, analysis, complex decisions

Best for:

Research tasks
Complex analysis
Tasks requiring external information
Multi-step reasoning

Model Tiers

Choose the right tier based on task complexity:

Tier	Use Case	Speed	Cost
Low	Simple classification, yes/no	Fastest	Lowest
Medium	Standard analysis, summarization	Balanced	Moderate
High	Complex reasoning, sophisticated analysis	Slower	Higher

Model Capabilities

Low Tier (Economy Processing)

Quick classification
Simple extraction
Binary decisions
High throughput tasks

Medium Tier (Analytical Reasoning)

Meeting summarization
Action item extraction
Content generation
Sentiment analysis

High Tier (Complex Reasoning)

Sophisticated analysis
Multi-factor decisions
Nuanced interpretation
Complex problem solving

Structured Outputs

AI nodes return typed, predictable outputs:

Return Type	Description	Example Use
`string`	Text response	Summaries, messages
`integer`	Whole number	Scores, counts
`boolean`	True/false	Classifications
`float`	Decimal number	Confidence scores
`string_list`	Array of strings	Action items, tags
`integer_list`	Array of numbers	Rankings, IDs

Prompt Structure

AI nodes use a messages array with roles:

text

[
  { role: "system", content: "Instructions for the AI" },
  { role: "user", content: "Context and data to process" }
]

Message Roles

Role	Purpose
`system`	Set AI behavior and instructions
`user`	Provide context and data
`assistant`	(Optional) Few-shot examples

Expression Languages

AI nodes use Liquid templates for dynamic content:

liquid

System: You are a meeting analyst.

User: Summarize this meeting.
Meeting: {{ json.meeting.title }}
Attendees: {{ json.meeting.attendees | map: "name" | join: ", " }}
Transcript: {{ json.callRecording.transcriptSummary }}

AI Workflow Patterns

Pattern: Analyze Then Act

text

[Load Data] → [AI: Analyze] → [If: condition] → [Action]

Pattern: Generate Content

text

[Load Data] → [AI: Generate message] → [Send message]

Pattern: Extract and Process

text

[Load Data] → [AI: Extract items] → [Select Many] → [Process each]

Pattern: Classify and Route

text

[Load Data] → [AI: Classify] → [If: urgent?]
                               ├─ Yes ─→ [Urgent path]
                               └─ No ──→ [Normal path]

Best Practices Summary

Practice	Impact
Choose appropriate model tier	Performance + Cost
Use transcriptSummary vs full transcript	Speed + Cost
Provide clear, specific instructions	Output quality
Use structured outputs	Reliability
Test with diverse data	Robustness

In This Section

Configuring AI Nodes - Setup and configuration
Prompt Engineering - Writing effective prompts
Structured Outputs - Working with typed responses
Model Selection - Choosing the right tier
Token Usage - Understanding and managing tokens