Explore a Custom Model

min read - Updated on Oct 28, 2026

The Explore page of a custom model shows the records the model learned from: every piece of text in its training set, with the label it carries. You can read the records one by one, filter them by label or by the words they contain, change a record's label, delete records, add new ones with New Record and download the whole training set with Download. A LABELS panel next to the records shows how the records are spread across the labels, so you can see at a glance which labels have few examples or unusually short ones.

Getting Ready

Custom models are available in the paid packages, not in Starter. See kimola.com/pricing.

You need a custom model created with a training dataset or with a Report, because these models have a training set to explore. Sign in to Kimola and open Models in the main menu.

A model created with a training dataset, or with a Report, learns what each label means from labeled examples. Its Explore page is where you check those examples after the model is built: whether every label has enough of them, whether they look like the feedback you will analyze, and whether some of them carry the wrong label. A model created with a prompt has no training set: it classifies feedback from the names and descriptions of its labels, which you see on its Overview page.

Browse Records

On the Models list, click View on the model's row to open its page, then open Explore. The page title shows the model's name followed by Explore, with the New Record and Download buttons on the right.

The records are listed as cards, 25 at a time; more load as you scroll down. Each card shows the text of the record, its label in the top right corner, the date the record was added (hover over it to see the full date and time), its length in characters, and a Delete button. A long text is cut after a few lines.

The LABELS panel on the right lists every label of the model with two bars:

Number of records shows the label's share of all the records in the training set, as a percentage.

Character length of records shows the label's share of the text in the training set, as a percentage, so you can compare how long each label's records are.

A model learns a label best when the label has enough examples and its examples look like the feedback it will classify. A label with a much smaller share of records than the others has fewer examples to learn from. A label whose share of the text is much smaller than its share of the records has shorter records than the rest, which often means they come from another source, such as one-word survey answers next to long support tickets. See Optimize Dataset-Based Custom Models for what to do about either.

Example

A mobile network operator trains a model on 1,200 support tickets with the labels Billing, Coverage, Roaming, Contract and Device. The LABELS panel shows Roaming with 3% of the records, far below the others, and Device with 18% of the records but only 6% of the text: its records turn out to be short app store comments mixed into the tickets.

Filter by Label

The two filters above the records narrow the list down, so you can read one label's examples on their own.

Open the label menu, which shows All Labels by default, and choose a label. The list then shows only the records that carry it. Choose All Labels again to see every record.

To find records by their words, type in the field that says "Type to filter records." The list updates when you stop typing or press Enter, and shows the records that contain what you typed. You can combine both filters, for example to look for the word "refund" among the records of one label.

Reading the records label by label is the quickest way to spot mislabeled examples and labels that overlap in meaning. Both lower the model's consistency, the percentage of semantically consistent pairs in its records.

Tip

Read each label's records on their own to spot mislabeled examples and labels that overlap in meaning.

Change or Delete a Record

When you find a record with the wrong label, you can correct it on the spot. Click the label on the record's card and choose the right label from the list. The card shows the new label, and the app says "Label updated as" followed by the label's name.

To remove a record from the training set, click Delete on its card. Kimola asks you to confirm with "This record will be deleted." Click I'm sure, delete! to delete it.

Note

A deleted record can't be brought back. If you may need it again, click Download first to keep a copy of the training set.

Add a Record

You can add examples to the training set one by one, for example to give a label with few records more examples to learn from. Click New Record at the top of the page. The New Record window opens with "Create a new record to add to the training set."

Type the text of the record in the text box. A record can be up to 500 characters long. Then open the menu below the text box, which says Choose a label, and choose one of the model's labels. A new record can only get a label the model already has. Click Create to add the record.

If the text box is empty, the window says "You must provide a content for the record." If no label is chosen, it says "You must provide a label for the record."

The window closes after each record unless you check Keep the window open before you click Create. With it checked, the window stays open and empties after each record, so you can add several records in a row. Click Close when you are done; the list of records then loads again with your new records in it.

Tip

To add many records at once, check Keep the window open: the window empties after each record so you can type the next one.

Use texts that look like the feedback the model will classify: written by customers, in the same language and of a similar length to the records already in the training set.

Download the Training Set

Download at the top of the page saves the model's training set to your computer as a spreadsheet file, with the records and their labels as they are now, including the records you added, changed or deleted on this page.

A downloaded training set is a convenient starting point when you want to build a new version of the model. Correct the file in a spreadsheet, then create the new version with Create AI Model › By Training Set. See Create a Custom Model with a Training Dataset and Optimize Dataset-Based Custom Models. It is also a backup of the training set before you delete records, or before you delete the model itself.

Note

Deleting a model deletes it with all its records. Download the training set first if you want to keep it. See Manage Custom Models.

To see how the model classifies new text after you have reviewed its records, use the test console on its Overview page. See Test a Custom Model.

Was this article helpful?

Tell us about your thoughts and experiences regarding the article.