View All Docs
Product overview
Account overview
Integrations
Transforms
Security
Hide Navigation
Integrations
Transforms
Security

Using chat to build and edit Flows

The fastest and easiest way to build in Parabola is simply by describing in plain language what you'd like to automate — from there, Parabola turns your intent into action.

The assistive sidebar is located to the left of your Flow canvas, and is available on any Flow that you have edit access to. Use the sidebar to chat with your Flow and convert your questions, instructions, and requirements directly into steps.

How it works

The primary purpose of the sidebar is to bring your ideas to life by actually adding steps to the canvas on your behalf. Through the sidebar, you can add new steps, edit existing steps, or delete steps that are no longer needed. It can change how arrows are connected, and re-arrange steps on your canvas.

The sidebar can also be used to chat about your Flow. If you don’t know what a Flow does, or how it is producing a certain result, just ask the sidebar! It can help summarize logic and direct you towards salient parts of your Flow.

In new Flows, use the sidebar to chat about your overall use case - what data you are going to need to access, what your ultimate goal is, and any considerations you already know about. You can talk through your use case to create a plan of action, and then let the chat take it from there.

Best practices: Writing effective prompts for Parabola's AI chat & Custom transform step

Overview

There are two main ways to write prompts in Parabola:

  • AI chat sidebar: Your AI teammate that helps you build and troubleshoot flows in plain language.
  • Custom transform step: A flexible step that translates plain language prompts into deterministic code that transforms your data.

Both tools are powered by large language models (LLMs), and both benefit from the same core principle: The more specific your prompt, the better your result.

Core Prompt Structure

Every strong prompt answers three basic questions:

  1. What’s the use case or goal?
    - Be specific about the outcome you're trying to achieve.
  2. What data are you working with?
    - Mention sources, formats, relevant columns, and assumptions.
  3. What should the output look like?
    - Describe final column names, data formats, and any visualizations or exports.

Example Prompts

✅ Good Prompts

Pulling data:

Help me build a sales report broken down by SKU using the last 30 days of data imported from Shopify.

Transforming data:

Filter out any sales where the “Gift card?” value is not blank.

Exporting data:

Export this report via email by sending an Excel file to sam@acme.com. Also, let’s create a data visualization showing sales by week.

Custom Transform:

Create a new column titled "WoW Change" based on the following logic:
- For each customer, calculate the # shipments from the previous week
- Compare it to the current week's shipments
- Calculate the % change
- Format as a percentage with 2 decimal points
Return all original columns

Best Practices for Writing Prompts

1. Be Specific, Not Vague

  • “Make a report.”
  • “Create a sales report by SKU, using the last 30 days of Shopify data.”

2. Reference Exact Columns

  • Use double quotes to refer to columns, e.g., "Order Date", "Gift card?".

3. Describe the Logic Clearly

Break down logic into steps:

- Add a column "Profit"
- Subtract "Cost" from "Revenue"
- Format as currency with 2 decimal places

4. Call Out Formats and Edge Cases

  • Specify currency, date, percentage, casing, etc.
  • If null values or duplicates need special handling, say so explicitly.

5. Use Clean Formatting

Structure your prompt like you’re writing an SOP:

  • Bullets or numbered lists
  • Line breaks between steps
  • Avoid paragraphs of dense text

Tips from the Parabola Team

💡 Talk like you're chatting with a colleague.
You don’t need to be an AI expert—just describe your process clearly, as if on Slack.

💡 Load your data early.
Parabola's chat performs best once your data is on the canvas and visible to the model.

💡 Think in bite-sized pieces.
Instead of writing one long prompt for your entire flow, iterate step by step. Prompt → Adjust → Build → Repeat.

💡 Keep prompts precise, not wordy.
You don’t need a wall of text—just be clear and focused.

💡 Use “Improve Prompt” when stuck.
The Custom Transform step has a built-in way to revise your prompt with AI help.

💡 Reset the chat when needed.
Stuck in a bad loop? Clear conversation history for a fresh start.

Prompt Optimization Checklist

When in doubt, check if your prompt includes:

✅ Clear use case or outcome
✅ Mention of data source(s)
✅ Specific columns referenced
✅ Desired format (currency, %, casing, date, etc.)
✅ Any edge cases or assumptions
✅ Clean formatting (bullets, newlines)

Final Word: When in Doubt, Try it Out

Prompts are iterative by nature. You won’t get it perfect on the first try—and that’s okay.

“You’re not going to break anything. Everything you do can be undone. Just describe your goal, try it out, and iterate as you go.”
— Isaac Wyatt, Automation Engineering Lead at Parabola

Product Overview  

Additional example prompts

Example prompts

  • Pulling data:
    • "Help me build a sales report broken down by SKU using the last 30 days of data imported from Shopify."
    • "I want to pull gift card data from Shopify and push data to a Google Sheet. Help me pull in data and set up the Flow." 
    • "I want to connect to the ShipHero API — help me pull in shipments." 
  • Transforming data:
    • "Filter out any sales where the 'Gift card?' value is not blank."
    • "Calculate the number of weekdays between the 'Order Date' and 'Ship Date' columns." 
    • "How do I classify all of my products by product type?" 
  • Exporting data:
    • "Export this report via email by sending an Excel file to jj@parabola.io."
    • "Create a data visualization showing sales by week."
    • "Let's send this data as a Slack message every time the Flow runs." 
  • Understanding Flows: 
    • "What is happening in this Flow?"
    • "Explain how the 'Calculate revenue by SKU' card works and break down the steps" 
    • "What columns are we using to calculate COGS?" 
  • Troubleshooting Flows: 
    • "Fix the error in the math step" 
    • "Why am I seeing an error in this step?" 

To share any feedback on your experience, use the thumbs up / thumbs down buttons in-app.

Best of luck building and let us know if you have any questions.