Flow Steps Overview

Steps are the building blocks of Flows in Cognipeer. Each step performs a specific function, from querying language models to executing database operations. This guide provides an overview of the different types of steps available for building your workflow.

Understanding Steps

Each step in a Flow:

• Accepts inputs either from Flow inputs or previous steps
• Performs a specific operation
• Produces outputs that can be used by subsequent steps

Steps are executed in sequence, creating a flow of operations that process data and produce results.

---

Available Step Types

Cognipeer offers a diverse range of step types to build powerful workflows:

AI and Language Model Steps

• Agent
• Ask to LLM
• Ask to Peer
• Guardrail
• Extract Data
• Intent Parser
• Speech to Text

Client-Side Execution

• Client Tool - Execute JavaScript functions on the client side

Data and Search Steps

• Search Knowledgebase

Web and Internet Steps

• HTTP Request
• Scrape Web Page

Workflow Control

• Condition
• Classifier
• Each
• While
• User Approval
• End Session
• Final

Interaction and UI Steps

• Add Widget to Message
• Show Form

Utilities

• Execute Node.js
• Check Regex
• Read File
• Read File From URL
• Convert To Markdown
• Markdown to PDF
• Chunk Content
• JSON to Excel
• Save File
• Save Text File
• Set State
• String to JSON
• JSON to String

Variable Manipulation

• Define Variable
• Set Variable
• Add Item to Array
• Add Items to Array
• Increment Variable
• Decrement Variable

Memory Management

• Save User Memory
• Search User Memory
• Delete User Memory

---

Combining Steps

The power of Flows comes from combining multiple steps into cohesive workflows. Here are some common patterns:

Sequential Processing

The simplest pattern is sequential processing, where steps are executed one after another, with each step using the output of the previous step.

Example:

1. HTTP Request step fetches data from an API
2. Execute Node.js step processes the data
3. Ask LLM step generates insights based on the processed data

Conditional Branching

Use the Condition step to branch flows. You can also pre-compute flags in Execute Node.js when needed.

Data Enrichment

Combine search and AI steps to enrich data:

1. Search Knowledgebase step retrieves relevant information
2. HTTP Request step fetches additional data from an external API
3. Ask LLM step combines and synthesizes the information

---

Best Practices for Using Steps

• Start Simple: Begin with a few steps and test thoroughly before adding complexity
• Use Descriptive Names: Give each step a clear, descriptive name that indicates its purpose
• Plan Your Data Flow: Consider how data will flow between steps and what transformations are needed
• Handle Errors: Use Node.js steps to implement error handling where appropriate
• Test Incrementally: Test your workflow after adding each new step to identify issues early

---

Next Steps

Explore the detailed documentation for each step type to learn about their specific configurations, inputs, and outputs:

• See the step reference links above for detailed pages.

By understanding the capabilities of each step type, you can design effective workflows that automate tasks, integrate systems, and leverage AI to solve business problems.