Skip to main content
Actions are the building blocks of your workflows. Choose between AI agents that can make intelligent decisions and direct tools for programmatic tasks.

​
Action Types

AI Agents
Intelligent assistants that can access multiple tools and make decisions. Capabilities:
  • Access to multiple integrations simultaneously
  • Natural language processing and decision-making
  • Context awareness across workflow steps
  • Ability to handle complex, multi-step tasks
Tools
Direct, programmatic actions for specific tasks. Capabilities:
  • Precise, deterministic operations
  • Fast execution for simple tasks
  • Direct API calls to connected apps
  • Structured data processing

​
Adding Actions to Workflows

Adding actions and tools to workflow
To add actions to your workflow:
  1. Click the blue + on any node or use the + in the flow dock
  2. Choose action type: Agent or Tool
  3. Configure settings: Select integrations and permissions
  4. Connect nodes: Link actions in logical sequence
  5. Add Exit node: Every workflow must end with an Exit node

​
Configuring AI Agents

Configuring agent toolkits and permissions

​
Toolkit Selection

Give your agent access to the tools it needs:
  • Select Toolkits: Choose integrations like Gmail, HubSpot, Google Sheets
  • Choose Connections: Select which specific accounts to use
  • Set Permissions: Configure read/write access levels

​
Approval Requirements

Control sensitive operations by requiring human approval:
  • Email Sending: Approve outbound messages
  • Data Deletion: Confirm destructive operations
  • Financial Actions: Verify payment processing
  • External Communications: Review customer-facing content

​
System Instructions

Adding system prompt and instructions to agent
Provide clear instructions for your agent: Example Instructions:
  • “Check my Gmail for any emails that require a response, and go ahead and respond for me”
  • “Analyze sales data from Google Sheets and create a summary report”
  • “Find new leads in HubSpot and send personalized LinkedIn connection requests”

​
Workflow Architecture

​
Sequential Actions

Create linear workflows where each action depends on the previous: 
Trigger → Agent (Data Collection) → Tool (Processing) → Agent (Analysis) → Exit

​
Parallel Processing

Execute multiple actions simultaneously for efficiency: 
Trigger → Agent → [Tool A, Tool B, Tool C] → Aggregation → Exit

​
Conditional Logic

Add decision points based on data or outcomes: 
Trigger → Agent → Decision Point → [Path A, Path B] → Exit

​
Complex Workflow Examples

Lead Processing Pipeline
  1. Apollo Tool: Find new prospects
  2. Agent: Analyze and qualify leads
  3. Google Sheets Tool: Add to CRM
  4. Gmail Agent: Send personalized outreach
  5. HubSpot Tool: Create follow-up tasks
Content Marketing Automation
  1. Web Search Tool: Find trending topics
  2. Agent: Generate content ideas
  3. Image Generation Tool: Create visuals
  4. LinkedIn Agent: Write and schedule posts
  5. Analytics Tool: Track performance
Customer Support Workflow
  1. Gmail Tool: Monitor support inbox
  2. Agent: Categorize and prioritize tickets
  3. Knowledge Base Tool: Find relevant solutions
  4. Agent: Draft personalized responses
  5. CRM Tool: Update customer records

​
Workflow Completion

Adding exit node to complete workflow

​
Exit Nodes

Every workflow must end with an Exit node to properly complete execution. You can use success exits for normal completion, error exits for failure scenarios, or conditional exits with different endpoints based on workflow outcomes.

​
Data Flow

Ensure data flows correctly between actions by mapping outputs from one action to the next and managing variables across workflow steps. Implement proper error handling to manage failed operations gracefully and maintain workflow stability.

​
Advanced Capabilities

Multi-agent collaboration allows multiple specialized agents to work on different aspects of complex tasks with coordinated decision-making. Dynamic tool selection enables agents to choose the best tools based on context and make runtime decisions for optimal efficiency. Build sophisticated 20+ step workflows for enterprise-grade automations that span multiple systems and implement complex business logic. These advanced workflows can handle intricate multi-system integrations and sophisticated decision trees.

​
Best Practices

Design agents with clear, specific instructions while granting only minimal necessary permissions and using approval workflows for sensitive actions. Test thoroughly before deployment to ensure reliable performance in production environments. Configure tools by choosing the most efficient option for each task while minimizing API calls and data transfers. Handle errors and edge cases proactively, and monitor performance regularly to identify optimization opportunities. Structure workflows to be focused and purposeful with descriptive names and comprehensive documentation. Plan for error scenarios and implement regular testing and maintenance schedules to ensure long-term reliability.

​
Next Steps

⌘I