Nodes
Complete reference for all node types in Workflow Builder
Nodes are processing units within a workflow that execute specific tasks. Each node has distinct configuration and functionality.
Start Node
The workflow entry point. Every workflow must begin with this node.
Configuration:
- Test Prompt: Input for testing workflow in preview mode
Rules:
- Cannot be deleted or renamed
- Only has output handle (right side)
- Required in every workflow
Agent Node
AI node with custom system prompt that processes input and generates responses.
Configuration:
- Node Name: Agent identification name
- Input Source Expression: CEL expression to retrieve input data (default:
input) - Instructions: System prompt for the agent
- Output Format: Text or JSON (with optional JSON Schema)
- Send Output to Chat: Toggle to send output to chat interface
- Load Conversation History: Toggle to load conversation history as context
- Save to Conversation History: Toggle to save responses to history
Usage:
- Set a descriptive Agent Node name
- Write instructions according to desired task
- If structured output is needed, select JSON format and define schema
- Enable conversation history if agent needs previous conversation context
Condition Node
Logic branching based on CEL expression evaluation. Has multiple output handles.
Configuration:
- Conditions: List of if/else-if conditions
- Case Name: Condition label (optional)
- Expression: CEL expression (example:
input.score > 80)
- Else Branch: Output for unmet conditions
Usage:
- Click "Add" to add new condition
- Write CEL expression (example:
input.status == "completed") - Connect each output handle to different target nodes
- Conditions are evaluated sequentially, first true condition executes
Expression Examples:
input.amount > 1000input.category == "urgent"input.score >= 80 && input.score < 90
Search Internet Node
Search for information from the internet using Google Custom Search API.
Configuration:
- Input Source Expression: CEL expression for search query
- Output Expression: Transform search results before passing forward
Usage:
- Ensure Google API Key and CX are configured in Settings
- Set expression for query (default: uses previous node output)
- Output contains search results and sources
Note: Requires valid API credentials.
Read Document Node
Read content from local files (PDF, DOCX).
Configuration:
- Input Source Expression: CEL expression for file path
- Output Expression: Transform file content before passing forward
Usage:
- Set expression to get file path (example:
input.file_path) - Node will extract content from file
- Supports formats: PDF, DOCX
Example: "C:\\Documents\\report.pdf" or input.document_path
Search Memories Node
Query from vector database memories using semantic search.
Configuration:
- Input Source Expression: CEL expression for search query
- Output Expression: Transform query results before passing forward
Usage:
- Set expression for query (default: previous node output)
- Node will search in LanceDB vector store
- Output contains relevant semantic search results
Note: Ensure data exists in memories database.
Call Command Node
Execute shell commands in the operating system.
Configuration:
- Input Source Expression: CEL expression for command to execute
Usage:
- Set expression for command (example:
"dir C:\\Users"orinput.command) - Command executes in PowerShell or Bash
Warning: Be cautious with dangerous commands. Use User Approval before this node for validation.
Call API Node
Call external HTTP APIs.
Configuration:
- Method: GET, POST, PUT, PATCH, DELETE
- URL: API endpoint
- Headers: Custom headers (JSON format)
- Body: Request body for POST/PUT/PATCH
- Output Expression: Transform API response
Usage:
- Select HTTP method
- Enter endpoint URL
- Set headers if needed (example:
{"Authorization": "Bearer token"}) - For POST/PUT, fill body with JSON data
- Output contains API response
Send Response Node
Send response to user through chat interface.
Configuration:
- Response Message: CEL expression for message to send
Usage:
- Set expression for message (example:
"Task completed: " + input.result) - Message is sent to chat interface
- Can concatenate text with data from previous node
Example: "Found " + string(input.count) + " results"
User Approval Node
Request user approval before continuing workflow. Dialog appears with Approve/Reject options.
Output Handles:
- Approved (green): Continue if user clicks Approve
- Rejected (red): Continue if user clicks Reject
Usage:
- Connect node before critical operations (delete, command, API call)
- When workflow reaches this node, approval dialog appears
- User reads context and chooses Approve or Reject
- Workflow continues according to user choice
Use Case: Confirmation before deleting files, executing dangerous commands, or API calls that modify data.
End Node
Terminate workflow. Workflow stops when reaching this node.
Rules:
- Cannot be deleted or renamed
- Only has input handle (left side)
- Multiple End nodes allowed for various exit points
Note Node
Workflow documentation. Not executed, for notes only.
Configuration:
- Note: Note text for documentation
Usage:
- Drag Note node to canvas
- Fill with notes, logic explanations, or reminders
- Node is not connected and does not affect execution
Use Case: Document complex logic, TODO notes, explanations for other developers.