> For the complete documentation index, see [llms.txt](https://docs.borndigital.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.borndigital.ai/digital-agent/conversation-flow/training-set/overview.md).
# Overview
The Training set screen: the intent list, adding intents (also from the intent library) and checking them with the confusion matrix.
Open **Design** › **Training set** in the project sidebar. The screen has two tabs: **Intent list** and **Intent validation**.
The training set holds the intents of the **Training set** type – intents recognised by a trained language model from example sentences (utterances). Keywords, Condition and AI intents are set directly in the Answer node and are not listed here. How recognition works is described in [Setting tab](/digital-agent/conversation-flow/setting-tab.md#intent-recognition-training-set-intents).
## Intent list
The intent list
| Column | Meaning |
| ---------------- | --------------------------------------------------------------- |
| **Intent name** | The name of the intent. |
| **Utterances** | How many example sentences the intent has. |
| **Used in node** | The Answer nodes that check this intent. |
| **Target node** | Where the conversation continues when the intent is recognised. |
Find an intent with **Search**. Click its name – or **Edit** in its **⋯** menu – to open it:
* add an utterance: type it in **Add utterance** and press **Enter**,
* remove an utterance with its trash icon,
* click **Back to intent list** when you are done.
**Delete** in the **⋯** menu removes the intent **immediately, without a confirmation**.
Aim for 5–15 different utterances per intent, written the way your customers really speak – see [Best practices for writing training set](/digital-agent/conversation-flow/training-set/best-practices-for-writing-training-set.md). Train the version again after you change the training set.
## Add an intent
1. Click **Add intent**.
2. Enter the name in **Name your new intent**.
3. Click **Add utterance** for every example sentence and type it.
4. Click **Add intent**.
**The intent library.** The upper part of the dialog offers ready-made intents in **Categories**, **Intents** and **Utterances**; the filter above them searches intent names and utterances. Drag the utterances you want into your new intent. The library exists for **Czech**, **Slovak** and **Polish** projects – for other languages it is empty.
The Add intent dialog (the library is empty for an English project)
## Intent validation
The **Intent validation** tab checks how well the trained model tells your intents apart, with a **confusion matrix**.
1. Train the version first – the matrix is calculated from a trained version.
2. Click **Generate confusion matrix**. In the dialog, **Generate matrix without generic intents** is ticked by default; tick **Check if confusion matrix already exists** to reuse the last result instead of calculating a new one.
3. The calculation can take **up to 15 minutes**. **Abort matrix generation** stops it.
The matrix compares every intent with every other one. The cells are coloured by score: **0.9 and more** is fine, **0.8** acceptable, **0.7** borderline, and **below 0.5** means the model often mixes up the two intents. Hover over a cell to see the utterances behind it (a click highlights its row and column), then make the utterances of the confused intents more distinct or merge the intents.
