ORCFLO Documentation

ORCFLO provides 8 node types to build your workflows. Each node type has specific configuration options that control its behavior. Understanding these nodes and their settings is essential for building effective automations.

Node Types

The 8 available node types each serve a specific purpose in your workflow.

Node Type	Purpose
Input	Defines the workflow entry point. Configure input parameters that users provide when running the workflow.
Output	Defines the workflow result. Maps internal step outputs to the final workflow output that users receive.
Note Card	Sticky notes for canvas annotations. Does not execute - purely for documentation and organizing your workflow visually.
LLM Step	The primary AI processing node. Configure prompts, select AI models (GPT-4o, Claude, etc.), set temperature, and define tools/functions.
Criteria Check	AI-powered condition evaluation using GPT-4o. Define natural language criteria and the AI determines if input meets the criteria.

LLM Step

The LLM Step is the workhorse of your workflow. It runs AI models to process data, generate content, classify inputs, and more.

Setting	Description	Options
AI Model	Select the model provider and specific model	GPT-4o, Claude 3.5, Gemini, etc.
Task Instructions	Instructions that define the AI's role and behavior	Text with variable support
Temperature	Controls randomness (lower = focused, higher = creative)	0.0 - 1.0

Example: LLM Step Configuration

Step Name: analyze_sentiment
AI Model: GPT-4o
Temperature: 0.3 (focused, consistent)

Task Instructions:
You are a sentiment analysis expert. Analyze the customer message
and respond with exactly one word: positive, negative, or neutral.

Temperature Guidelines

Use low temperature (0.1-0.3) for classification, extraction, and factual tasks. Use higher temperature (0.7-0.9) for creative writing and brainstorming.

Input & Output Nodes

Input and Output nodes define your workflow's interface - what data comes in and what results go out.

Input Node Settings

Field Name: Identifier for the input parameter
Field Type: String, number, boolean, object, array
Required: Whether the field must be provided

Output Node Settings

Output Mapping: Map step outputs to final result fields
Output Schema: Define the structure of workflow output

Control Flow Nodes

Control flow nodes (Criteria Check) direct how your workflow executes based on conditions.

Criteria Check Configuration

Input Value: The data to evaluate (variable reference)
Criteria: Natural language description of the condition (e.g., "The customer is expressing frustration or anger")
AI Model: Uses GPT-4o for evaluation

Approval Settings

ORCFLO supports human-in-the-loop workflows with step and tool approval settings. Require human approval before executing sensitive operations.

Setting	Description
Step Approval	When enabled, the workflow pauses before executing this step and waits for human approval. Users can review the step inputs and approve or reject.
Tool Approval	For LLM Steps with tools/functions, require approval before the AI invokes specific tools. Useful for actions with side effects like sending emails or modifying databases.

Persistent Approvals

Approval settings can be configured to persist across executions. Once approved, a step or tool call can be auto-approved for future runs.

Data Flow Between Nodes

Data flows automatically through connections in your workflow. When you connect nodes, the output from one step becomes available to downstream steps. Simply write plain text in your task instructions and reference what you need naturally.

Connect nodes to establish data flow between them
Downstream nodes automatically receive upstream outputs
Write plain text in task instructions - the AI understands context
All upstream data is accessible, not just directly connected nodes

Natural Language Instructions

Simply describe what you want in your task instructions. The AI model receives the context from connected upstream nodes automatically.

Best Practices

Use descriptive step IDs: Name steps by function: "analyze_sentiment" not "step_2"
Document with Note Cards: Add sticky notes to explain complex logic on the canvas
Use Criteria Check for nuance: When conditions require understanding context, use AI evaluation

Common Mistakes to Avoid

Mistake	Fix
Using an LLM Step for complex conditions	Use Criteria Check for nuanced evaluations
Overly complex task instructions	Keep instructions clear and focused - let data flow through connections
No approval on destructive actions	Enable step or tool approval for operations with side effects

Node Configuration