Create, edit, and debug your queries faster
AI-written movie scripts? Never. AI-written queries? Yes, please.
The joy and beauty of a good movie script comes from the unique insights and experiences of the writer, not from the writer’s prowess at the keyboard. Data-driven decisions don’t come from knowing SQL minutia but from incisive questions from curious people. We understand that data analysts often face challenges when writing queries, hindering their ability to focus on what really drives value and we want to change that.
At Dune, we are committed to unleashing the full power of Language Model-based Learning (LLM) to make crypto data accessible to all. Our goal is to empower our users to delve into what matters most, abstracting away the tedious aspects of their work. To achieve this vision, we are thrilled to announce the launch of three LLM-powered products: Create Wand, Edit Wand, and Debug.
Like Aphrodite being born of sea foam, your SQL query will emerge fully formed and ready to deliver insights. Just kidding.
This is our most ambitious LLM product: generating an entire query from a natural language prompt. You will be happy (or sad) to know that you are not being completely replaced by robots. This product is most useful for creating queries and getting over the hump of a blank query editor.
With the Create Wand, we aim to streamline the query writing process, by simplifying the initial stages of query development, analysts can swiftly get to the heart of their questions and accelerate the exploration of valuable insights.
For each natural language question, we’ll search for the most semantically similar examples from our library of Dune-specific few shots, include all table schemas for prompted tables, and auto-debug any incorrect syntax generated by the LLM. This is a product that we intend to continuously improve as we learn more about how wizards write prompts and build the most accessible blockchain natural language querying engine.
Here is an example of the Create Wand in action:
The Edit Wand allows you to take an existing query (bootstrapped or bespoke) and write an edit prompt to modify it. For a lot of data work, the difficulty is finding the right tables and fields to filter on. After that, defining the summary statistics, window functions, or reformatting with CTEs is not a particularly rewarding process. Why not focus on getting the core logic of the data you want to analyze and let an LLM define the CROSS JOIN conditions for you?
See our docs for some inspiration on how to use the feature.
AI-written jokes? Not even once.
Syntax errors are the worst, especially when you are learning a new dialect. That’s why we’ve launched a feature that automatically debugs your query. We soft-launched this feature a few weeks ago, and we’re consistently fixing ~90% of syntax errors submitted. Never again stare at your screen again searching for a missing comma.
Here's Debug in action:
We're excited to be making progress towards our LLM roadmap that we announced back in March. As an experimental phase, we are excited to offer these three LLM features— Wand Create, Wand Edit, and Debug to our community. Your input is valuable to us. We invite you to provide us with your feedback, suggestions, and any issues you encounter. Your insights will help us fine-tune and optimize the tools.
LFG to making crypto data accessible.
Solidus Labs demystifies the nuances of A-A wash trading in the DeFi ecosystem.
Building a startup is a walk on the wild side. Survival can’t be taken for granted.
The best work happens via collaboration, which is why we’re focused on building tools that simplify data workstreams and open up new possibilities for collaboration and automation. Today we’re excited to launch a suite of new features to usher in a new era for teams building on Dune.
The revolution will not be reported quarterly.
Celebrating Dune wizards. Dashboards, stories, alpha dropped in your inbox weekly.