elvex flows support different node types, each designed for specific purposes within your workflow. This reference explains what each node type does, when to use it, and its key configuration options.
Basic Step
The foundation of elvex flows - a customizable LLM prompt with configurable tools and datasources.
When to use it: For most standard flow operations where you need the LLM to process information, generate content, or perform analysis.
Key configuration:
Instructions (required): Specific directions for what the step should accomplish
Provider and model: Use flow defaults or specify custom settings
Tools: Enable capabilities like web search, webpage content retrieval, spreadsheet analysis, or SQL queries
Datasources: Connect relevant knowledge bases or files
Result validation: "Manager" that checks output quality (enabled by default)
Testing: Use the Test tab to try the step with sample inputs and see outputs before running the full flow.
Ask an Assistant
A node that leverages existing assistants within your flow.
When to use it: When you want to incorporate specialized assistants you've already built rather than recreating their functionality.
Key benefits:
Reuses existing assistants with their specialized configurations
Maintains consistent assistant behavior across different contexts
Avoids duplicating setup work
Typical uses: Incorporating domain-specific assistants, content generation assistants, or analysis assistants into larger workflows.
Decision
A branching node that evaluates input and directs flow based on defined criteria.
When to use it: When your workflow needs to take different paths depending on specific conditions.
Key configuration:
Decision instructions (required): Criteria for making the decision (e.g., "Is this a sandwich?")
Options: Define two or more possible paths (e.g., "Yes" and "No")
Option criteria: Optional explanation for when each option should be chosen
No match option: Optional path for when none of the defined options apply
Example:
Instructions: "Determine if the text describes a food item"
Option 1: "Yes" - When input describes something edible
Option 2: "No" - When input doesn't describe food
Call an existing Flow
A node that embeds another flow as a step within your current flow.
When to use it: When you want to create modular, reusable workflow components or break complex processes into manageable parts.
Key benefits:
Creates modular, reusable workflow components
Simplifies maintenance (update once, changes reflect everywhere)
Helps organize complex processes into logical components
Considerations:
Be mindful of performance when nesting multiple flows
Understand dependency relationships between parent and child flows
Ensure proper testing of the complete flow structure
Loop
A container node that processes multiple items through the same series of steps automatically.
When to use it: When you need to process lists, spreadsheets, or text data by running each item through the same workflow steps.
Key configuration
Splitting strategy (required): How to divide input into individual items
Comma: Split by commas (CSV-style lists)
Rows: Split by rows (spreadsheets) or newlines (text)
Sentences: Split by sentence-ending punctuation
Paragraphs: Split by paragraph breaks (double newlines)
Data source (optional): Specific file or data source to use as input. If not selected, uses attached files or previous step output
Maximum iterations (default: 20): Maximum number of items to process (range: 1-20)
Two output modes
Loop output (standard):
Location: Bottom-right handle
Processes all items in parallel (fast)
Combines all results into single output
Use for: Independent items that don't need context from each other
Iteration output (feedback):
Location: Bottom-left handle (yellow)
Processes items sequentially (slower)
Each iteration receives all previous results as context
Use for: Items that build upon previous results
Visual indicator: Yellow "Reduce" chip when configured
Typical uses
Analyze each row in a spreadsheet independently
Translate documents sentence by sentence
Generate personalized content for lists of recipients
Iteratively refine content where each step builds on previous results
Process batch operations with consistent logic
Considerations
Each iteration counts as a separate AI call (monitor costs)
Maximum 20 iterations per loop execution
5-minute timeout per iteration
Feedback loops process sequentially (slower but enables context)
All steps inside the loop must be connected
Combine these node types to create sophisticated workflows that handle complex business processes while maintaining clarity and reusability.