Get the code, not the result, to keep your real data private

The prompt

When you need an LLM to solve a task involving your data, such as analysis, transformation, or a calculation, ask it to write the code that does the job, not to run the computation itself and hand back the answer. Run that code yourself, locally, on your real data.

If the model needs to see the shape of your data to write correct code, give it a small sample or generate synthetic or fake data with the same structure instead of your real records.

Expected result

A working script or function you run yourself, instead of your raw data being pasted into a chat window and processed by the model directly.

Why it works

Asking for code instead of a computed result means your actual data never has to leave your machine. You control exactly what the model sees, and you can hand it fake data shaped like the real thing when it needs an example to work from.