Warning: This document is for an old version of Rasa X.

Annotate Conversations

Annotating conversations in Rasa X allows you to easily and efficiently update your training data based on real conversations.

There are a number of different ways that you can improve your assistant through annotating conversations in Rasa X:

Video Tutorial

Check out this video tutorial for a more detailed explanation of annotating conversations with Rasa X.


The conversations view lets you inspect all the conversations users have had with your assistant on any channel.


Add conversations

When you first start Rasa X, you will have no conversations. If you already have an existing Rasa assistant, you can view all of its conversations here by importing them into Rasa X.

To get new conversations, you will need to either invite users to test your bot or connect a custom channel (e.g. Slack, Twilio, etc) and have them chat with your assistant.

You can learn more about the supported messaging and voice channels as well as how you can connect your own, which all can be used with Rasa X, in the Rasa Docs.

Conversations that you have with your assistant via ‘Talk to your bot’ or ‘Share your bot’ will not show up here. However, if you chat with your assistant via the ‘Share your bot’ link in an Incognito window, you can see what your testers will see and those conversations will show up here.

Correct NLU or Core mistakes

If you click on a message, you can correct any Rasa NLU mistakes. If you correct its intent or label an entity by highlighting it in the message at the top of the right sidebar, that NLU example will be added to your training data.


If NLU was correct, but Core selected the wrong action, you can copy the story up to that point into Interactive Learning mode and correct the behaviour. Once you correct the mistake and click ‘Save’, a new story will be added to the stories used to train your model. Remember, you will have to re-train your model before these changes come into effect.

Flag a message

If you are not sure how to correct something, you can always flag a message and revisit it later. A flag will show up next to the conversation, so that you can easily know which conversations need revisiting.


Filter conversations

There is also a filter button in the top right that lets you narrow down which conversations you want to view (e.g. by date; conversations with flagged messages; conversation length; actions, entities, and intents that should be present).


Talk to your bot

In the Talk to your bot tab, you can have a conversation with your assistant and generate a story. If you click ‘Save’, that example story will be added to the stories used to train your model.


Interactive Learning

In the Talk to your bot tab, you can also switch the mode from ‘Talk’ to ‘Interactive Learning’. Here you can provide feedback to your bot while you talk to it and generate a story. If you click ‘Save’, that example story will be added to the stories used to train your model.


Next to each message, you can also click ‘Correct this’, which will let you correct a certain input or action and continue the conversation from the corrected event.

You can learn more about Interactive Learning in the Rasa Docs.

NLU Training

Label NLU examples from conversations

The Annotate New Data tab under NLU Training will auto-populate with user inputs and their predicted intents from the conversations page. Highlighting words in the sentence allows you to label entities, while choosing from the dropdown of intents or typing in a new intent will let you label its intent. Once you click ‘Mark Correct’, then that example will move from ‘Annotate New Data’ to ‘Training Data’. You will need to retrain your model in order for this new training data to improve your NLU performance.


Create new NLU training data

You can also add new NLU examples in this tab by clicking the plus icon.


Filter new NLU examples

If you want to only look at examples that have been labeled as certain intents, you can filter by those intents.