Support AI-assisted SQL generation via ClickHouse client integration

[niyue]

Use Case

Starting from ClickHouse 25.7, ClickHouse Client and clickhouse-local include AI-powered functionality that converts natural language descriptions into SQL queries. This feature allows users to describe their data requirements in plain text, which the system then translates into corresponding SQL statements. [1]

chdb could support similar capability without too much effort. This functionality is particularly valuable during exploratory data analysis. In environments such as Jupyter notebooks, an AI model can quickly interpret the table schema, generate representative sample data, and streamline a variety of routine tasks.

Describe the solution you'd like

The ClickHouse client already implements this capability in C++, and chdb could leverage the same code path and expose it through its language bindings, such as Python. This avoids code duplication, and any future improvements made by the ClickHouse team would automatically benefit chdb as well.

References

[1] AI-powered SQL generation, https://clickhouse.com/docs/use-cases/AI/ai-powered-sql-generation

[niyue]

I’ve submitted a PR (#440) that introduces this support. Any feedback would be greatly appreciated. Thanks.

[wudidapaopao]

ClickHouse uses the ?? prefix to denote the AI-powered SQL generation feature. Since chDB operates in a Python environment, could we directly add two types of interfaces? One could be named generate_sql, which directly generates SQL via a prompt. The other could be named ask or chat, which generates SQL, executes it directly, and returns the results to the user (with Pandas DataFrame as the default return format).

[niyue]

That sounds reasonable. I consider this option previously but was hesitate because it would introduce new API which requires other bindings to implement and I would like to keep the change scope small initially.

I will follow your advice to revise the code later. Thanks.