Creating Workflows

Build AI-powered workflows using the visual drag-and-drop builder.

Workflow Builder

The visual workflow builder provides a canvas for creating complex automations by connecting nodes. Each node represents an action, trigger, or control flow operation.

Builder Components:

• Node Palette: Left sidebar with available nodes (collapsible)
• Canvas: Drag-and-drop workspace with zoom/pan controls
• Config Panel: Right sidebar for node configuration
• Execution Panel: Bottom panel for testing and debugging
• AI Chat: AI assistant for workflow generation and help

Building Your First Workflow

Basic Workflow Structure

Every workflow needs:

1. Trigger Node: Start point (Start, Schedule, Webhook, or App trigger)
2. Action Nodes: Processing steps (AI agents, tools, API calls)
3. End Node: Termination point (optional but recommended)

Creating a Workflow

1. Click Create Workflow from workflows list
2. Drag nodes from palette to canvas
3. Connect nodes by dragging from output handle to input handle
4. Configure each node by clicking it
5. Click Save to store workflow
6. Click Run to test execution

Trigger Nodes

Trigger nodes initiate workflow execution.

Start Node

Purpose: Manual trigger for on-demand execution

Configuration: No configuration required

Use Cases:

• Ad-hoc workflows
• Testing and debugging
• Manual data processing

Schedule Trigger

Purpose: Time-based workflow execution

Configuration:

• Cron Expression: Schedule pattern (e.g., 0 9 * * * for daily at 9 AM)
• Timezone: Execution timezone (default: UTC)
• Enabled: Toggle trigger on/off

Common Patterns:

• 0 9 * * * - Daily at 9 AM
• 0 9 * * 1 - Every Monday at 9 AM
• 0 */6 * * * - Every 6 hours
• 0 0 1 * * - First day of each month

Use Cases:

• Daily reports
• Scheduled data sync
• Recurring notifications

Webhook Trigger

Purpose: HTTP webhook-based execution

Configuration:

• Webhook URL: Auto-generated endpoint
• Method: HTTP method (POST, GET, PUT, DELETE)
• Authentication: API key or none

Use Cases:

• External system integration
• Event-driven workflows
• Third-party notifications

Accessing Webhook Data: Use input1 or trigger.body in downstream nodes to access webhook payload.

App Trigger

Purpose: Application event-based execution

Configuration:

• App: Select integrated application
• Trigger: Choose specific event
• Parameters: Event-specific settings

Use Cases:

• CRM updates
• Form submissions
• Database changes

AI & Model Nodes

AI-powered processing nodes for intelligent workflows.

Agent Node

Purpose: AI agent with tool-calling capabilities

Configuration:

• Model: AI model (e.g., openai/gpt-5)
• Temperature: Creativity (0-1, default: 0.7)
• Max Tokens: Response length limit (default: 2000)
• System Prompt: Agent instructions
• Max Steps: Maximum tool calls (default: 5)

Use Cases:

• Complex reasoning tasks
• Multi-step problem solving
• Tool orchestration

Accessing Inputs: Reference previous node outputs using input1, input2, etc.

Prompt Node

Purpose: Simple text generation without tools

Configuration:

• Content: Prompt template

Template Variables:

• {input1}, {input2}: Previous node outputs
• {variable_name}: Custom variables

Use Cases:

• Content generation
• Text transformation
• Summarization

Embedding Model Node

Purpose: Convert text to vector embeddings

Configuration:

• Model: Embedding model (e.g., openai/text-embedding-3-small)
• Dimensions: Vector size (default: 1536)

Use Cases:

• Semantic search
• Document clustering
• Similarity matching

Structured Output Node

Purpose: Extract structured data from AI responses

Configuration:

• Schema Name: Output schema identifier
• Mode: object or array

Use Cases:

• Data extraction
• Form filling
• Structured data generation

Image Generation Node

Purpose: Generate images from text prompts

Configuration:

• Model: Image model (e.g., gemini-2.5-flash-image)
• Aspect Ratio: Image dimensions (1:1, 16:9, 9:16)
• Output Format: png or jpeg

Use Cases:

• Marketing content
• Visual assets
• Creative generation

Audio Node

Purpose: Text-to-speech conversion

Configuration:

• Model: TTS model (e.g., openai/tts-1)
• Voice: Voice selection (alloy, echo, fable, onyx, nova, shimmer)
• Speed: Playback speed (0.25-4.0, default: 1.0)

Use Cases:

• Voice notifications
• Content accessibility
• Audio content generation

Published Agent Node

Purpose: Reuse existing configured agents

Configuration:

• Agent: Select published agent
• User Message: Input prompt
• Agent settings inherited from published agent

Use Cases:

• Agent reusability
• Standardized workflows
• Team collaboration

Action Nodes

Execute operations and external integrations.

Tool Node

Purpose: Custom tool definition

Configuration:

• Name: Tool identifier
• Description: Tool purpose

Use Cases:

• Custom integrations
• Specialized operations

HTTP Request Node

Purpose: Make HTTP API calls

Configuration:

• URL: API endpoint
• Method: GET, POST, PUT, DELETE, PATCH
• Headers: Request headers (JSON)
• Body: Request payload (JSON)
• Authentication: Bearer token or none

Use Cases:

• REST API integration
• External data fetching
• Webhook delivery

Accessing Response: Output available as result in downstream nodes.

Web Scrape Node

Purpose: Extract content from web pages

Configuration:

• URL: Target web page
• Selector: CSS selector (optional)

Use Cases:

• Content aggregation
• Data collection
• Monitoring

Retrieval Node

Purpose: Semantic search over documents

Configuration:

• Documents: JSON array of documents
• Top K: Number of results (default: 3)
• Model: Embedding model

Use Cases:

• Knowledge base search
• Document Q&A
• Context retrieval

Example Documents:

["Vercel is a frontend cloud.", "Next.js is a React framework."]

Composio Action Node

Purpose: Execute Composio integrations

Configuration:

• App: Composio app
• Action: Specific action
• Parameters: Action inputs

Use Cases:

• Third-party integrations
• Multi-platform automation
• SaaS connectivity

JavaScript Node

Purpose: Execute custom JavaScript code

Configuration:

• Code: JavaScript function

Available Variables:

• input1, input2, etc.: Previous node outputs
• return: Function return value

Example:

// Transform input to uppercase
return input1?.toUpperCase?.() ?? input1

Use Cases:

• Data transformation
• Custom logic
• Calculations

Python Node

Purpose: Execute Python code

Configuration:

• Code: Python script
• Packages: Required packages (comma-separated)

Available Variables:

• inputs: List of all inputs
• input1, input2, etc.: Individual inputs
• print(): Output to logs

Example:

# Access inputs
print('input1:', input1)
result = input1.upper()

Use Cases:

• Data science workflows
• ML model inference
• Complex calculations

Control Flow Nodes

Manage workflow execution paths.

Conditional Node

Purpose: Branch execution based on condition

Configuration:

• Condition: JavaScript expression (e.g., input1 === 'value')

Outputs:

• True Handle: Condition passes
• False Handle: Condition fails

Use Cases:

• Decision logic
• Error handling
• Dynamic routing

Condition Examples:

• input1 > 100 - Numeric comparison
• input1.includes('error') - String matching
• input1?.status === 'success' - Object property check

Fork Node

Purpose: Split workflow into parallel paths

Configuration:

• Branches: Number of parallel outputs

Use Cases:

• Parallel processing
• Multiple simultaneous actions
• Independent operations

Note: Fork creates multiple execution paths. Use Join node to merge results.

Join Node

Purpose: Merge multiple parallel paths

Configuration:

• Branches: Number of inputs to wait for
• Merge Strategy:
  • array: Combine all inputs as array
  • object: Merge as object properties

Use Cases:

• Aggregating parallel results
• Synchronization points
• Data combination

End Node

Purpose: Workflow termination point

Configuration: No configuration required

Use Cases:

• Explicit workflow completion
• Multiple end points
• Success/failure paths

Connection Rules

Handle Constraints

Regular Nodes:

• Input: One incoming connection only
• Output: One outgoing connection only

Special Nodes:

• Fork/Conditional: Multiple outputs allowed
• Join: Multiple inputs allowed

Connection Errors

“Node can only have one outgoing connection” Solution: Use Fork node to split into multiple paths

“Node can only have one incoming connection” Solution: Use Join node to merge multiple paths

Workflow Execution

Testing Workflows

1. Click Run button in toolbar
2. Execution panel opens at bottom
3. View real-time node status:
   • Idle: Not yet executed
   • Running: Currently processing
   • Completed: Finished successfully
   • Error: Failed with error
4. Check execution logs for debugging
5. Review node outputs

Execution Flow

• Nodes execute in sequence following connections
• Parallel paths (from Fork) execute concurrently
• Join waits for all inputs before proceeding
• Errors stop execution at failed node

Debug Tips

• Check node outputs in execution panel
• Review console logs for errors
• Test individual nodes before connecting
• Use simple workflows first, add complexity gradually

Workflow Management

Saving Workflows

New Workflow:

1. Click Save
2. Enter workflow name and description
3. Workflow created with status: DRAFT

Existing Workflow:

• Auto-saves on changes (debounced)
• Manual save via Save button
• Unsaved changes indicated by * in title

Workflow Status

DRAFT: Development mode, not active ACTIVE: Enabled, triggers registered PAUSED: Temporarily disabled ARCHIVED: Deprecated, read-only

Activation

Requirements:

• Workflow must be saved
• Valid trigger node required
• All nodes properly configured

Activate: Click Activate button to enable triggers

Deactivate: Click Deactivate to pause execution

AI Workflow Assistant

Access AI assistant via Sparkles button (bottom-right).

Capabilities:

• Generate complete workflows from description
• Add nodes to existing workflows
• Explain workflow logic
• Suggest improvements
• Debug issues

Example Prompts:

• “Create a daily report workflow”
• “Add error handling to this workflow”
• “How do I schedule this for weekdays only?”

Best Practices

Workflow Design:

• Start simple, add complexity incrementally
• Use descriptive node labels
• Test frequently during development
• Add comments using JavaScript nodes

Error Handling:

• Add conditional checks for error states
• Use try-catch in JavaScript/Python nodes
• Provide fallback paths
• Log errors for debugging

Performance:

• Minimize sequential steps
• Use parallel execution (Fork) where possible
• Optimize API calls
• Cache expensive operations

Maintainability:

• Document complex logic
• Use published agents for reusability
• Keep workflows focused on single purpose
• Version workflows before major changes

Sample Workflows

Load pre-built workflows via Sample Workflows dropdown:

• Daily email reports
• Data processing pipelines
• API integration examples
• Multi-step AI workflows

Customize samples to fit your needs.

Keyboard Shortcuts

• Tab: Navigate between nodes
• Delete: Remove selected node
• Cmd/Ctrl + C: Copy node
• Cmd/Ctrl + V: Paste node
• Cmd/Ctrl + Z: Undo
• Cmd/Ctrl + S: Save workflow

Next Steps

• Review Workflows Overview for management
• Explore AI Agents for agent creation
• Check Tasks for simple automations