approximatelabs/sketch

Leverages data sketches and summary statistics to provide LLM context about dataframe contents, enabling accurate code generation for pandas workflows without IDE plugins. Offers three complementary interfaces—`.ask()` for data exploration, `.howto()` for code generation, and `.apply()` for row-level transformations—with flexible backend options including OpenAI, local Hugging Face models (MPT-7B, StarCoder), or Sketch's remote endpoint.

2,284 stars and 353 monthly downloads. Used by 1 other package. No commits in the last 6 months. Available on PyPI.