Variable Fill

Variables control how data moves through your agent workflows. Every tool parameter requires a fill method that determines how the variable receives its value during execution.

Understanding Variable Fill Methods

Variable fill methods define how data populates into tool parameters at runtime. Choosing the correct fill method ensures smooth data flow and accurate results.

Linked

Connect outputs from previous nodes to current tool inputs

AI Fill

Agent automatically extracts values from context and conversation

Static

Fixed values that remain constant across all executions

User Fill

Values provided by users at runtime or via API

From Memory

Retrieve values from agent’s long-term memory and past interactions

Fill Methods

Linked

Connect outputs from previous nodes to inputs of current nodes. This is the primary method for chaining workflow steps together. How It Works:

Previous node executes and generates output variables
Current node references those outputs by name
Values automatically flow when current node executes
No manual mapping required after initial setup

When to Use:

Passing data between sequential workflow steps
Using results from one tool as inputs to another
Building multi-step processing pipelines
Maintaining data continuity across nodes

Configuration:

Node A Output: customer_record
Node B Input: customer_data
Fill Method: Linked → Node A: customer_record

Note: Variables can also link non-sequentially (e.g., Step 2 output → Step 5’s input) via linked variables. You’re not limited to connecting only adjacent nodes. Example Workflow:

Node 1: Retrieve Customer (Integration Connector)
├─ Output: customer_record {id, name, email, status}

Node 2: Analyze Customer Status (Custom GPT)
├─ Input: customer_info
└─ Fill Method: Linked → Node 1: customer_record

Selecting Fill Method: When configuring a variable, you’ll see the fill method dropdown with all 5 options:

Best Practices:

Link outputs don’t have to be to the next node. Encourage matching keys within steps for accurate mapping
Use descriptive variable names to track data provenance
Verify data types match between output and input
Check linked parameters appear in node execution logs

AI Fill

AI automatically extracts or infers values from available context (conversation history, documents, memory). How It Works:

Agent analyzes all available context
AI extracts relevant information based on parameter description
Value populates automatically without explicit mapping
Falls back to null if data not found

When to Use:

Extracting information from unstructured text
Inferring values from conversation context
Processing document content
Classifying or categorizing based on context

Configuration:

Parameter: customer_intent
Fill Method: AI Fill
Description: "Classify customer inquiry as: Sales, Support, Billing, or Other"
Context Source: Conversation history, uploaded documents

Example:

User Message: "I need help with my invoice from last month"

Tool Parameter: inquiry_category
Fill Method: AI Fill
Description: "Category of customer inquiry"

AI Populates: inquiry_category = "Billing"

Best Practices:

Write clear parameter descriptions to guide AI extraction
Enable memory lookup for richer context
Provide examples in description for complex extractions
Test with diverse inputs to verify accuracy

Static

Fixed values that never change during workflow execution. Set once during configuration and remain constant. How It Works:

Define value during tool configuration
Value remains identical across all executions
No runtime computation or lookup required
Fastest fill method (no processing overhead)

When to Use:

API endpoints and URLs
Configuration parameters
Default values and fallbacks
Business rules and thresholds
System identifiers

Configuration:

Parameter: notification_email
Fill Method: Static
Value: "[email protected]"

Common Use Cases:

Parameter	Static Value	Purpose
api_endpoint	”https://api.company.com/v2”	API base URL
default_priority	”medium”	Fallback priority level
company_name	”Acme Corporation”	Standard company identifier
max_retries	3	Error handling threshold
email_footer	”© 2025 Company”	Standard email signature

Best Practices:

Use for values that truly never change
Document why value is static (avoid future confusion)
Review static values periodically for relevance
Consider User Fill if value may need runtime override

User Fill

Values provided explicitly when workflow executes through the user interface. How It Works:

Workflow pauses at node requiring user fill
User provides value via interface
Validation rules check input (if configured)
Execution continues with provided value

When to Use:

Form submissions and user inputs
Runtime configuration during workflow execution
Dynamic values that vary by execution
Human intervention points in workflows

Configuration:

Parameter: order_id
Fill Method: User Fill
Required: true
Data Type: string
Validation: Must match format "ORD-XXXXX"
Description: "Order ID to process"

Example Scenario: User Input During Execution:

Workflow pauses at approval step

User is prompted to provide:
- approval_decision: "approve" or "reject"
- approval_notes: "Approved for Q1 budget allocation"
- next_reviewer: "[email protected]"

Workflow continues with provided values

Best Practices:

Mark critical parameters as required
Add validation rules to prevent errors
Provide clear descriptions for UI display
Set sensible default values when appropriate

From Memory

Retrieve values from agent’s memory system, including past interactions, learned preferences, and stored context. How It Works:

Agent searches memory for relevant information
Retrieves data based on parameter description and memory keys
Uses semantic search to find best match
Returns most recent or relevant value found

When to Use:

Accessing conversation history
Retrieving user preferences
Loading previous interaction data
Building on past context

Configuration:

Parameter: customer_preferences
Fill Method: From Memory
Memory Lookup: Enabled
Context Window: Last 10 interactions
Description: "Customer's communication preferences and history"

Best Practices:

Enable memory lookup in agent settings
Use specific memory keys for faster retrieval
Set appropriate context windows
Combine with AI Fill for intelligent defaults

Choosing the Right Fill Method

Decision Framework:

Does value come from previous node output?
└─> Use Linked

Is value always the same?
└─> Use Static

Does user provide value at runtime?
└─> Use User Fill

Should AI extract from context/documents?
└─> Use AI Fill

Should agent retrieve from memory?
└─> Use From Memory

Common Patterns:

Tool Type	Typical Fill Methods
First node in flow	User Fill, Static, From Memory
Middle processing nodes	Linked, AI Fill
Integration nodes	Linked, Static
Final output nodes	Linked, AI Fill

Variable Data Types

All fill methods support multiple data types: Primitive Types:

String: Text values
Number: Integers and decimals
Boolean: true/false values
Date: Timestamps and dates

Complex Types:

Object: Nested key-value structures
Array: Lists of values
JSON: Structured data objects

Example Type Configurations:

{
  "customer_email": {
    "type": "string",
    "fillMethod": "user_fill",
    "validation": "email"
  },
  "order_total": {
    "type": "number",
    "fillMethod": "linked",
    "source": "calculate_total.output.amount"
  },
  "is_priority": {
    "type": "boolean",
    "fillMethod": "ai_fill",
    "description": "Whether request is high priority"
  },
  "line_items": {
    "type": "array",
    "fillMethod": "linked",
    "source": "extract_items.output.items"
  }
}

Validation and Error Handling

Ensure data quality with validation rules: Validation Types: Format Validation:

Email addresses: [email protected]
Phone numbers: +1-XXX-XXX-XXXX
URLs: https://example.com
Custom regex patterns

Range Validation:

Minimum/maximum values
String length limits
Array size constraints
Date ranges

Required vs Optional:

Required: Workflow fails if missing
Optional: Continues with null/default

Example Validation:

{
  "email": {
    "fillMethod": "user_fill",
    "type": "string",
    "required": true,
    "validation": "email"
  },
  "age": {
    "fillMethod": "ai_fill",
    "type": "number",
    "min": 0,
    "max": 120
  },
  "tags": {
    "fillMethod": "linked",
    "type": "array",
    "minItems": 1,
    "maxItems": 10
  }
}

Linked Records (Integration-Specific)

When working with database integrations (Airtable, databases), linked record fields require specific formatting. Airtable Linked Records: Linked record fields must be arrays of record IDs, even for single links:

{
  "linked_contact": {
    "fillMethod": "linked",
    "type": "array",
    "value": ["recABC123"]
  }
}

Common Pattern:

Node 1: Find Contact Record
├─ Output: contact_id = "recABC123"

Node 2: Update Invoice (Airtable)
├─ Field: Linked Contact
└─ Value: [contact_id]  // Must be array format

Multiple Linked Records:

{
  "related_projects": {
    "fillMethod": "linked",
    "type": "array",
    "value": ["recXYZ789", "recABC123", "recDEF456"]
  }
}

Troubleshooting

AI Fill not populating variable

Common Causes:

Insufficient context available
Parameter description too vague
Memory lookup not enabled
Required data not in conversation

Solutions:

Enhance parameter description with examples
Enable memory lookup in agent settings
Verify context contains required information
Use User Fill as fallback for critical data

Linked variable showing undefined

Common Causes:

Source node hasn’t executed yet
Output variable name mismatch
Source node execution failed
Data type incompatibility

Solutions:

Verify source node executed successfully
Check exact output variable name
Review node execution logs
Confirm data types match between nodes

User Fill validation failing

Common Causes:

Input doesn’t match validation rules
Data type mismatch
Required field not provided
Custom regex pattern too strict

Solutions:

Review validation rules and test cases
Provide clear error messages
Add examples to parameter description
Adjust validation patterns if needed

From Memory returns empty

Common Causes:

Memory lookup not enabled
No relevant data in memory
Memory key doesn’t exist
Context window too narrow

Solutions:

Enable memory lookup in agent configuration
Expand context window size
Use specific memory keys
Combine with AI Fill as fallback

Airtable linked records failing

Common Causes:

Record ID not in array format
Invalid record ID format
Record doesn’t exist
Permission issues

Solutions:

Ensure IDs are in array: ["recXXX"]
Verify record exists in Airtable
Check field is configured as linked record type
Confirm API permissions allow linking

Best Practices

Fill Method Selection:

Default to Linked for chaining nodes
Use AI Fill for unstructured data extraction
Reserve Static for truly constant values
Enable User Fill only when runtime input needed
Leverage From Memory for personalization

Variable Naming:

Use descriptive, consistent names
Follow naming conventions (snake_case recommended)
Include data type hints in complex objects
Avoid abbreviations unless standard

Performance Optimization:

Minimize AI Fill usage (slower than other methods)
Use Static where possible (fastest)
Cache From Memory results
Batch User Fill prompts together

Error Handling:

Set required fields appropriately
Provide meaningful validation messages
Add fallback values where sensible
Test with edge cases and missing data

Next Steps

Selecting Tools

Choose and configure tools that use variable fill

Creating Flows

Design workflows that leverage variable fill methods

Integrations

Configure integration-specific variable patterns and connectors

Custom Integrations

Build custom API connections for advanced integration patterns

Getting Started

Building Agents

Running & Operations

Observability & Analytics

Resources

Understanding Variable Fill Methods

Linked

AI Fill

Static

User Fill

From Memory

Fill Methods

Linked

AI Fill

Static

User Fill

From Memory

Choosing the Right Fill Method

Variable Data Types

Validation and Error Handling

Linked Records (Integration-Specific)

Troubleshooting

Best Practices

Next Steps

Selecting Tools

Creating Flows

Integrations

Custom Integrations

Getting Started

Building Agents

Running & Operations

Observability & Analytics

Resources

​Understanding Variable Fill Methods

Linked

AI Fill

Static

User Fill

From Memory

​Fill Methods

​Linked

​AI Fill

​Static

​User Fill

​From Memory

​Choosing the Right Fill Method

​Variable Data Types

​Validation and Error Handling

​Linked Records (Integration-Specific)

​Troubleshooting

​Best Practices

​Next Steps

Selecting Tools

Creating Flows

Integrations

Custom Integrations

Understanding Variable Fill Methods

Fill Methods

Linked

AI Fill

Static

User Fill

From Memory

Choosing the Right Fill Method

Variable Data Types

Validation and Error Handling

Linked Records (Integration-Specific)

Troubleshooting

Best Practices

Next Steps