Quickstart with PromptQL
Introduction
This guide is designed to give you a high-level overview of PromptQL and how you can use it to make better decisions about your business and automate tasks; all without any coding knowledge required.
Before continuing, you'll need a Hasura Cloud account.
If you have your own project, you can apply the information below to it. Otherwise, we have a
collection of sandboxes which allow you to try PromptQL and get a feel for the experience. Below,
we'll use the sandbox-gtm example.
If you're looking to build a new project from scratch, check out these guides.
Guide
Open your project
Each PromptQL project can be accessed via a unique URL. You can access all your projects from the PromptQL Console.
When you navigate to a project, you're dropped into its playground; this is the chat interface that allows you to ask questions about your data and create automations.
Talk to your data
Primarily, you'll interact with PromptQL in one of two ways: either by asking questions, or creating automations.
Ask a complex question
Using the sandbox-gtm project, we're going to ask the following question:
Which enterprise deals are at risk of slipping this quarter based on engagement patterns?
While this question seems simple, it's complex because it requires defining what qualifies as an enterprise deal, identifying signals that indicate a deal is at risk—such as low engagement or delayed progress—and analyzing various engagement patterns, including sales activities, buyer signals, and product usage, all in comparison to historical and successful deal benchmarks.
If you're using your own project, start by asking a similarly strategic question relevant to your data. Otherwise, click here to be dropped into a thread replay.
Query plan
When you ask a question, PromptQL will respond with a query plan. This is a step-by-step breakdown in plain language of how it will process your request, including any data retrieval, filtering, or AI operations it will perform.
It helps make PromptQL's approach transparent by explaining what it's going to do before it does it, including any assumptions it's making about the data or process.
Take a look at the query plan generated for your question and see how PromptQL takes a complex question and breaks it down into a series of steps, just like a human would.
You can edit a query plan by clicking the pencil next to a query plan.
You have domain-specific expertise which you can share with PromptQL. When you modify a query plan, PromptQL remembers and self-corrects for future threads!
Execution
The query plan is then executed against your data.
PromptQL pulls the necessary data and uses AI to help make sense of it. Behind the scenes, it fetches information from your database and organizes it using Python, so you can focus on the insights, not the code.
If you're curious, expand the execution in your project and see the code PromptQL is executing against your datasource. And, just like the query plan, if you see something that should be different, let PromptQL know!
Artifacts
Typically, the execution of a query plan will result in an artifact.
Artifacts are structured outputs like tables, text, or visualizations that store the processed data or analysis results, which can be referenced later in the conversation, exported, or used to start new threads for further analysis.
We have a more in-depth guide for making decisions with PromptQL; check it out here.
Talking with PromptQL is like talking to a human. This means it's an iterative process wherein you ask a question, it responds, and you build on that response by refining, clarifying, or asking follow-up questions. The more context you provide, the better the answers become.
Give it a task to automate
Using the same sandbox project, we'll ask PromptQL to generate an automation for us:
I need an automation to help me identify risky opportunities in my sales pipeline. I want to input a date range and get back a list of opportunities that need attention. Specifically, I want to know which deals:
- Haven't had any recent activity
- Are missing important contact relationships
- Show low engagement from the customer
- Are taking longer than expected to close
Can you create something that gives me a risk score for each opportunity and suggests specific actions I should take? I'd like the results in a format that makes it easy for me to prioritize which deals need immediate attention.
This task-based prompt asks PromptQL to build an automation: a reusable workflow that runs against your data and gives you consistent, structured results. Rather than just answering a one-time question, automations are designed to help you monitor, score, or act on key insights regularly.
In this example, PromptQL will identify the relevant data, define what makes a deal “risky,” build a scoring system to prioritize those deals, and suggest actions—all packaged in a format that’s easy to review and reuse. Just like with query plans, PromptQL explains each step and lets you modify the logic, so you stay in control of the process.
If you're using your own project, give PromptQL a similar task to automate. Otherwise, click here to be dropped into a thread replay.
We also have a dedicated guide to building your own automations available here.
Next steps
Now that you've had your first set of conversations with PromptQL, we recommend checking out these docs around its core capabilities.