I use DuckDB for queries and Visidata for quick inspections.
Between those two, I can work with not only CSVs, but also JSON and Parquet files (which are blazing fast -- CSVs are good for human readability and editability, but they're horrendous for queries).
CLI CSV tools pop up every now and then, but there's too many of them and I feel that my use cases are sufficiently addressed with only 2 tools.
If you use jupyter check out what I'm building with Buckaroo. The aim is to have a table viewer that does the right/obvious thing for most data (formatting, sorting, downsampling large datasets, summary statistics, histograms). all customizable. Supports pandas and polars.
Between those two, I can work with not only CSVs, but also JSON and Parquet files (which are blazing fast -- CSVs are good for human readability and editability, but they're horrendous for queries).
CLI CSV tools pop up every now and then, but there's too many of them and I feel that my use cases are sufficiently addressed with only 2 tools.