Dune Introduces “Wand” Suite and Debug: LLM‑Driven Query Tools for Crypto Analysts
San Francisco – March 4 2026 – Dune, the leading analytics platform for blockchain data, announced the public beta of three new features that harness large‑language‑model (LLM) capabilities to streamline SQL‑based querying. The tools – Create Wand, Edit Wand, and Debug – are designed to lower the technical barrier for analysts, allowing them to focus on insight generation rather than on the intricacies of query syntax.
Why LLM‑assisted querying matters
Data teams working with on‑chain metrics frequently encounter two bottlenecks: (1) translating a business question into a correct SQL expression, and (2) troubleshooting syntax errors that interrupt exploratory workflows. While recent advances in generative AI have made natural‑language interfaces feasible, many platforms still require a deep familiarity with the underlying schema. Dune’s newest suite aims to bridge that gap by injecting LLM intelligence directly into the query‑building lifecycle.
Feature roundup
| Feature | Core functionality | How it works | Expected benefit |
|---|---|---|---|
| Create Wand | Generates a complete SQL statement from a plain‑language prompt. | The model searches Dune’s internal repository of “few‑shot” examples that match the user’s intent, injects the relevant table definitions, and auto‑corrects any syntactic flaws before returning the query. | Cuts the time spent staring at an empty editor, especially for users new to blockchain schemas. |
| Edit Wand | Refines an existing query based on a natural‑language instruction. | Users feed an original query and a short edit directive (e.g., “add a 30‑day moving average”). The LLM revises the statement, handling joins, window functions, or CTE restructuring automatically. | Enables rapid iteration without manually re‑writing complex sub‑queries. |
| Debug | Detects and repairs SQL syntax errors on the fly. | Once a query is submitted, the system parses the error, runs the LLM‑driven fixer, and presents a corrected version, reporting the change made. Early beta data shows a fix‑rate of roughly 90 % for reported issues. | Reduces the “debug loop” that often stalls analysts, especially those still learning the dialect of Dune’s blockchain‑specific tables. |
The UI integrates each function into Dune’s existing query editor, preserving familiar workflows while offering an optional “wand” button for each step. Visual demos released by Dune illustrate the tools in action: a natural‑language request about “total ETH transferred by top‑10 wallets in the last week” instantly produces a ready‑to‑run script, while a malformed query is corrected within seconds.
Industry perspective
The move comes as more blockchain analytics firms experiment with AI‑augmented data access. “The biggest friction point for DeFi researchers is translating high‑level strategy questions into precise, performant queries,” says Laura Chen, senior analyst at CoinMetrics. “If Dune can reliably generate correct SQL from everyday language, it not only speeds up research but also democratizes access for non‑technical participants.”
Other platforms, such as The Graph and Flipside Crypto, have hinted at AI‑assisted query generation, but Dune is the first to ship a full‑stack, production‑grade set of tools that combine generation, editing, and debugging under a single brand.
Potential impact on the DeFi ecosystem
- Accelerated insight cycles – Faster query creation means analysts can iterate on hypotheses more quickly, potentially shortening the time from data discovery to market‑relevant action.
- Lowered entry barriers – New entrants, including venture funds and media outlets, can extract on‑chain metrics without a dedicated data engineer.
- Higher query quality – The auto‑debug feature reduces human error, leading to cleaner data pipelines and more reproducible research.
- Competitive pressure – Competitors may need to match or differentiate their AI capabilities, spurring further innovation in the space.
However, the reliance on LLMs also raises concerns about model hallucination—producing plausible but inaccurate queries. Dune mitigates this by grounding generation in a curated set of few‑shot examples and by automatically validating syntax, but semantic correctness will still require human oversight.
Key takeaways
- Three LLM‑powered tools (Create Wand, Edit Wand, Debug) now in public beta on Dune, aimed at simplifying blockchain SQL queries.
- Create Wand turns natural‑language questions into fully formed queries; Edit Wand modifies existing scripts on demand; Debug automatically fixes syntax errors with a reported 90 % success rate.
- The suite is built on Dune’s internal library of example queries and table schemas, ensuring that generated code is aligned with the platform’s data model.
- Early feedback loops are encouraged through a dedicated feedback portal, allowing Dune to iterate rapidly based on community usage.
- Industry analysts see the tools as a step toward broader democratization of on‑chain analytics, though vigilance around model accuracy remains essential.
Looking ahead
Dune framed the launch as part of a broader “LLM roadmap” disclosed earlier this year. The company indicated that the Wand suite will continue to evolve, incorporating more sophisticated prompt engineering, deeper schema awareness, and expanded support for multi‑chain environments. As the beta progresses, user adoption metrics and further refinements will likely shape the next generation of AI‑driven analytics tools in the DeFi sector.
For analysts eager to test the features, Dune has opened the Wand suite to all existing account holders, with a feedback form to collect suggestions and report any quirks. The initiative marks a clear signal that the crypto‑data industry is embracing generative AI not as a novelty, but as a core productivity enhancer.
— Prepared for CryptoPulse, March 4 2026
Source: https://dune.com/blog/dune-llms
