Paste a table into the chat, get a summary
Copy a chunk out of a spreadsheet into a message and get the read straight back: what it is, what stands out, what looks wrong.
What the scenario looks like
- Execute + Data
- LLM
Sometimes a table is not for a report but for working out what is in it. Uploading it somewhere for a single question is not worth it, and building a pivot takes longer than asking.
What is inside
- A pasted table — the entry. The data arrives inside the message.
- Read and summarise — a model whose prompt fixes three parts: what the data is, three or four observations with numbers, and one line on what looks suspicious.
- The summary — the exit.
The third part is the most useful. Gaps, outliers and duplicates are the last thing a human notices and the first thing that ruins any conclusion drawn later.
What to set up for yourself
- The prompt. Your three parts may differ: “compare with the previous period”, or “name the top five and the tail”.
- The input format. Anything that copies works: CSV, a chunk of a spreadsheet, a plain list of lines. Demanding a specific format is pointless — the model copes, and a person will not remember the rules.
What to connect
Nothing: no files, no connections, no base. The ceiling is the size of a message: a hundred rows go through easily, ten thousand will fit neither into a message nor into a sensible price per turn. For big tables there is “Ask your own table in plain words”, where a node reads the file.