Add columns

Create a static column for values you set yourself, or a dynamic one that computes them, from the same panel.

Every column starts in the same panel, whether it holds values you type in or values a method computes for you. Open it, pick a type, name the column, and either save it right away or test it first.

From the dataset’s Data tab, click Add Column. The Add Columns panel opens with a filter on the left: All, Static Columns, or Dynamic Columns. Pick a filter (or search by name), then click the type you want. A static column creates immediately once you name it; a dynamic column opens a fuller form for the method’s settings, and lets you test it before you commit.

Add a static column

Static columns are the fastest path: pick a data type, name it, and it’s in the grid ready to fill in.

Pick a type

Filter to Static Columns and click the data type you want, for example Text. The full set of types, and what each one stores, is in Limits & Data Types.

Name it and create

A small panel opens with a Column name field. Name it reviewer_notes and click Add Column.

The column appears in the grid right away, empty, ready for you to fill in cell by cell.

Add a dynamic column

Dynamic columns point at a method instead of holding values you type. The example here uses Classification, which reads another column’s text and sorts each row into one of the labels you define. Run Prompt is a dynamic column too, but it gets its own walkthrough in Run a prompt on every row; every other method is cataloged in Dynamic column methods.

Choose the method

Filter to Dynamic Columns and click Classification.

Name it and pick the source column

Name the column query_topic, then select the column to classify: user_query.

Define the labels

Add each label the model can choose from, for example Billing, Technical, and Account.

Pick a model and set concurrency

Choose the model that runs the classification, and set how many rows it processes at once.

Test, then create

Click Test to preview the labels it would assign, without saving anything yet. Once it looks right, click Create New Column.

Create New Column starts Classification on every row, filling query_topic in as it works through the dataset.

Validation you’ll hit

Column names cap at 255 characters
A name that’s already used in this dataset is rejected
Two columns in the same request can’t share a name, which only comes up when you add more than one column at once, for example through the SDK

Questions & Discussion