nebulaflow

Basic Workflow Examples

This guide provides simple workflow examples to help you understand the core concepts of NebulaFlow.

Example 1: Simple Greeting Workflow

Goal

Generate a greeting message using an LLM and display it.

Workflow Structure

Start → LLM Node → CLI Node (Shell) → End

Step-by-Step Setup

1. Add Start Node
   • Drag from palette to canvas
   • No configuration needed
2. Add LLM Node
   • Drag from palette
   • Configure:
     • Model: openai/gpt-4o (or any available model)
     • System Prompt: “You are a friendly assistant.”
     • User Prompt: “Generate a warm greeting for a new user.”
     • Temperature: 0.7
3. Add CLI Node
   • Drag from palette
   • Configure:
     • Content: echo "${1}" (the full command line; ${1} refers to the first parent output)
     • (Optional) Mode: command (default)
     • (Optional) Shell: bash (default)
4. Add End Node
   • Drag from palette
   • No configuration needed
5. Connect Nodes
   • Connect Start → LLM
   • Connect LLM → CLI
   • Connect CLI → End

Execution

Click the Run button. The LLM will generate a greeting, and the CLI node will print it.

Example 2: File Processing Workflow

Goal

Read a file, process it with an LLM, and save the result.

Workflow Structure

Start → CLI (Read) → LLM (Process) → CLI (Write) → End

Step-by-Step Setup

1. Start Node - Default

2. CLI Node (Read)
   • Content: cat input.txt (the full command line)
   • (Optional) Mode: command (default)
   • (Optional) Shell: bash (default)
3. LLM Node
   • Model: openai/gpt-4o (or any available model)
   • System Prompt: “You are a text processor. Summarize the following text concisely.”
   • User Prompt: ${1} (the output from CLI Read)
   • Temperature: 0.5
4. CLI Node (Write)
   • Content: echo "${1}" > summary.txt (where ${1} is the LLM output)
   • (Optional) Mode: command (default)
   • (Optional) Shell: bash (default)
5. End Node - Default

Execution

The workflow reads a file, generates a summary, and saves it to a new file.

Example 3: Conditional Processing

Goal

Process data differently based on its content.

Workflow Structure

Start → LLM (Analyze) → If/Else → [True] LLM (Process A)
                                   → [False] LLM (Process B)
                                    → End

Step-by-Step Setup

1. Start Node - Default

2. LLM Node (Analyze)
   • Model: openai/gpt-4o (or any available model)
   • System Prompt: “Analyze the following text and determine if it’s positive or negative.”
   • User Prompt: ${1} (the output from Start Node)
   • Temperature: 0.3
3. If/Else Node
   • Content: ${1} == "positive" (where ${1} is the output from LLM Analyze)
   • True Branch: Connect to LLM Node A
   • False Branch: Connect to LLM Node B
4. LLM Node A (Positive)
   • System Prompt: “Generate an encouraging response.”
   • User Prompt: ${1} (the output from LLM Analyze)
5. LLM Node B (Negative)
   • System Prompt: “Generate a supportive response.”
   • User Prompt: ${1} (the output from LLM Analyze)
6. End Node - Default

Execution

The workflow analyzes input text and routes it to different processing paths based on sentiment.

Example 4: Looping Workflow

Goal

Process multiple items in a list.

Workflow Structure

Start → Loop Start → LLM (Process) → CLI (Save) → Loop End → End

Step-by-Step Setup

1. Start Node
   • Output: (optional) can be used to override iteration count
   • Example: 3 (if using override)
2. Loop Start Node
   • Iterations: 3 (fixed number of iterations)
   • Loop Variable: i (available as ${i} in loop body)
   • Loop Mode: fixed
   • (Optional) Override iterations: connect a parent node to the special input port
3. LLM Node
   • System Prompt: “Generate a greeting for iteration ${i}.”
   • User Prompt: Iteration ${i} (or any content)
   • Temperature: 0.7
4. CLI Node
   • Content: echo "${1}" >> results.txt (where ${1} is the LLM output)
   • (Optional) Mode: command (default)
5. Loop End Node
   • No configuration needed
6. End Node - Default

Execution

The workflow runs the loop body 3 times, each time using the iteration index ${i}. The LLM generates a greeting, and the CLI appends it to a file.

Example 5: API Integration

Goal

Fetch data from an API and process it.

Workflow Structure

Start → CLI (Fetch) → LLM (Analyze) → CLI (Send) → End

Step-by-Step Setup

1. Start Node - Default

2. CLI Node (Fetch)
   • Content: curl https://api.example.com/data
   • (Optional) Mode: command (default)
3. LLM Node
   • System Prompt: “Analyze the following JSON data and extract the main insights.”
   • User Prompt: ${1} (the output from CLI Fetch)
   • Temperature: 0.5
4. CLI Node (Send)
   • Content: curl -X POST -H "Content-Type: application/json" -d '${1}' https://api.example.com/insights
   • (Optional) Mode: command (default)
   • (Optional) Shell: bash (default)
5. End Node - Default

Execution

The workflow fetches data, analyzes it with an LLM, and sends the insights to another API using a CLI node with curl.

Common Patterns

Sequential Processing

A → B → C → D

Simple linear flow where each node processes the output of the previous one.

Branching

A → B → C
    ↘ D → E

Conditional branching based on node output.

Parallel Execution

A → B → C
A → D → E

Multiple paths from the same node (requires careful data handling).

Looping

A → Loop Start → B → C → Loop End

Repeating operations on multiple items.

Best Practices

Start Simple

• Begin with 2-3 nodes
• Test each node individually
• Add complexity gradually

Test Incrementally

• Run the workflow after each node addition
• Check intermediate outputs
• Use Preview nodes for debugging

Handle Errors

• Add condition nodes for error checking
• Use try-catch patterns where possible
• Log errors for troubleshooting

Optimize Performance

• Batch operations when possible
• Use appropriate delays
• Cache results when reused

Next Steps

• Advanced Workflows - More complex patterns
• Integration Examples - External service integration
• Node Types - Detailed node documentation

This site is open source. Improve this page.