Have you seen Ibis[1]? It's a dataframe API that translates calls to it into various backends, including Polars and DuckDB. I've messed around with it a little for cases where data engineering transforms had to use pyspark but I wanted to do exploratory analysis in an environment that didn't have pyspark.
I think it would be better to have one of each of the main AIs / coding tools running instead of all the same model and tool. That way you can keep tabs on latest developments while doing your work. Also if you use multiple providers / tools, you can take advantage of more free credits that way, because some providers have a few free credits each month.