Denodo Assistant for VQL Shell¶
The Denodo Assistant has some features integrated into the VQL shell. To use these features, type your request directly into the Denodo Assistant panel, and the Assistant will automatically select the appropriate feature. Alternatively, you can access the features directly from the panel using the new Denodo Assistant commands:
In this panel you can also:
Include custom user instructions that will apply to all interactions with the Assistant in the VQL Shell integration. To do this, click the gear icon on the Denodo Assistant panel and enter your instructions.
Select a subset of databases to limit your Denodo Assistant requests to (applies to /generate and /metadata). When one or more databases are selected, only views from those databases will be considered by the Denodo Assistant.
Diagnose Query¶
The feature Diagnose query analyzes a given query, providing both a corrected version and an explanation of the modifications made.
Generate Query From Natural Language¶
The feature Generate query from natural language analyzes a plain language query written by the user and provides a corresponding VQL query along with an explanation of what the query does.
Metadata Query¶
The feature Metadata Query analyzes a natural language input to provide a list of the most relevant views, presenting an explanation for the selection along with a table with links for easy access.
Improving Query Accuracy¶
Here are some considerations to take into account to improve the accuracy of the generated query and explanation:
Sample Data¶
To generate accurate VQL, the Denodo Assistant can analyze a small sample of a view’s actual data alongside its metadata. This helps the AI resolve format ambiguities that aren’t defined in the field descriptions.
For example, if you ask the Denodo Assistant “generate a query that lists all employees from California”, the AI must
decide if the filter should be WHERE state = 'California' or WHERE state = 'CA'.
By checking a data sample, the AI can see the actual format used and generate the correct clause immediately.
This data sampling also resolves other common ambiguities:
Date formats: Differentiating between 2025-09-02 and 09/02/2025.
Categorical Codes: Determining if a status is a string ‘Shipped’ or a numerical code like 3.
Enabling sample data ensures more reliable and precise VQL queries on the first attempt. You can activate this feature by enabling the Use sample data option in the Denodo Assistant configuration.
Descriptions¶
Descriptions for views, fields, and tags provide the essential business context the AI needs. While the Denodo Assistant is powerful, it does not inherently know your organization’s specific terminology.
Clear, human-readable descriptions are critical for accuracy when technical names are not self-explanatory (e.g., c_bal, ord_dt, or is_act). For instance, a field named status is ambiguous, but a description like “Customer account status: A for Active, I for Inactive, S for Suspended” provides the precise detail the AI needs to answer questions correctly.
By providing this precise detail, you directly improve the AI’s ability to interpret your data and generate more intelligent, accurate VQL.