A language model that reads the question, not one that writes the query.
Text-to-SQL chat tools ask a language model to write a database query from a question. Corlogic uses the language model only to interpret the question; the SQL is compiled deterministically from an approved business model.
The model writes the query.
The model is given the question and a description of the tables, and it writes SQL. This is quick to set up and works well for exploration by people who can read and check the query.
A query can run and still be wrong.
A generated query can run without error and still be wrong: a join that duplicates rows, or a measure such as revenue calculated in a way the business does not use. Nothing in the answer signals the problem unless someone reads the SQL and knows the data.
Interpretation by the model, compilation by rules.
| Text-to-SQL chat tool | Corlogic | |
|---|---|---|
| Role of the language model | Writes the SQL. | Interprets the question as a structured request against the approved StarEngine model. |
| Where business definitions live | In the prompt, or in the model's guess. | In StarEngine, which records which definition applies and how it is calculated. |
| How the SQL is produced | Generated, so it can vary from one run to the next. | Compiled deterministically from the defined business logic, as exact, read-only SQL. |
| Unsupported questions | May be answered anyway. | Refused with a specific reason. |
| Access control | Depends on how the tool is connected. | Role-based permissions, fail-closed row-level security and a read-only, least-privilege query role. |
For analysts who read the query.
For analysts exploring data they know well, a text-to-SQL assistant is a useful shortcut. The difference matters when answers go to people who will act on them without reading the query.
More on how Corlogic answers questions: Corlogic IQ and StarEngine.
Query ExecutionStart with a recurring process that is harder to coordinate than it should be.
Talk to us about where context gets lost, decisions stall, or follow-up breaks down across people and systems.
Talk to us