Celeste flow tips

​

Overview

Celeste is your AI assistant for flow building. Instead of manually creating nodes one by one, describe what you want in natural language and let Celeste build it for you! What Celeste Can Do:

• Build entire flows from natural language descriptions
• Add or modify existing nodes in existing flows
• Configure complex formulas and conditions
• Suggest optimizations for flow performance
• Explain what a flow does

​

How Celeste Works

Click the Celeste icon at the bottom center of your screen to open the chat. How It Works:

1. You describe what you want in natural language
2. Celeste analyzes your request and your current flow
3. Celeste generates nodes and connections
4. You review and accept the changes

​

Building and Modifying Flows with Celeste

​

Example 1: Creating a Simple Flow

Your Prompt:

“Create a flow that filters sales over $1000, groups by region, and calculates total sales per region”

Celeste Response: Celeste creates a Flow with these nodes:

1. Get Data node (you select the data source)
2. Transform node that filters sales > 1000, groups by region, and sums sales
3. Save Table node

​

Example 2: Adding to Existing Flow

Your Prompt:

“Sort the results by total sales, highest first, in the @Weekly Sales Orders flow”

Celeste Response: Celeste updates the Transform node to sort by:

• Sort field: total_sales (the aggregated field)
• Sort direction: Descending

​

Example 3: Complex Formula

Your Prompt:

“Add a calculated field called profit_margin that’s (revenue - cost) / revenue * 100 in the @Weekly Sales Orders flow”

Celeste Response: Celeste adds a calculated field in the Transform node:

• New field name: profit_margin
• Formula: ([revenue] - [cost]) / [revenue] * 100

​

Effective Celeste Prompts

​

Good Prompts

• Specific: “Filter customers where status = ‘active’ AND created_date > ‘2024-01-01’”
• Clear Goal: “Calculate average order value by customer segment”
• Field Names: “Join orders table with customers table on customer_id”
• Step-by-Step: “First filter sales by date, then group by product, then sort by total descending”

​

Less Effective Prompts

• Too Vague: “Clean my data”
• Missing Details: “Calculate something with sales” (what calculation?)
• Ambiguous: “Fix the flow” (what’s wrong?)

Tip: If Celeste asks clarifying questions, answer them to get better results.

​

More Prompt Examples

A few more requests that work well, across common operations:

• Filter: “Show only rows where region is ‘West’ or ‘East’ and sales_amount > 500”
• Transform: “Add a full_name field by combining first_name and last_name with a space”
• Aggregate: “Calculate total revenue and average order value by product category”
• Join: “Left join with the products table on product_id, keeping all rows from orders”
• Complex logic: “Add a customer_tier field: ‘Gold’ if total_purchases > 10000, ‘Silver’ if > 5000, otherwise ‘Bronze‘“

​

Reviewing AI-Generated Flows

Always review Celeste’s changes, as LLMs can make mistakes:

​

Check Configuration

1. Verify Field Names: Ensure Celeste used the correct field names from your data
2. Check Formulas: Review calculated fields for accuracy
3. Validate Filters: Confirm filter conditions match your intent
4. Test Logic: Run the flow and preview results

​

Celeste Flow Explanations

Celeste can explain existing flows: Your Prompt:

“Explain what the @Weekly Sales Orders flow does”

Celeste Response:

“This flow:

1. Loads sales data from a Get Data node
2. Filters to sales over $1,000
3. Groups by region and calculates total sales
4. Sorts regions by total sales (highest first)
5. Saves the top 10 regions to a table

Use Case: Identify your best-performing sales regions”

​

Best Practices