In May, we launched Rasa X—tooling designed specifically to enable Rasa developers to create better conversational AI assistants. It’s been a great success. With more than 700 developers tuning in for our announcement webinar and over 75,000 downloads since it was launched, thousands of developers have started to use Rasa X to learn from real conversations.
Many of you have taken the time to submit bugs on Github and share your ideas on how to make it better. With this release, we focused on polishing and improvements based on this feedback, and we are planning more such releases in the coming months.
Why Rasa X?
At Rasa, we are focused on empowering developers to build and deploy the best possible contextual assistants. We believe that this means we need to ship applied research that pushes the capabilities of Rasa (e.g knowledge bases, response retrieval models, Transformer Embedding Dialogue Policy). But we also know that algorithms alone won't solve conversational AI, which is why Rasa X was built to help you collect more training data.
The first version of Rasa X was a great step in this direction. As tooling that enables developers to improve their assistants based on conversations with real users, it solves a big problem around how to collect high quality data and use it to improve assistants, so they can more reliably and successfully chat with users.
However, this first version did not have support for all Rasa features, and that caused friction. Rasa X 0.21 was released today and includes improved support for the things you love in Rasa as well as updates to the features you already use in Rasa X.
What’s new in 0.21?
A key focus of this latest release was to ensure important features you use when building with Rasa are included in Rasa X. This is a big step toward ensuring that all of the richness of Rasa will be available in Rasa X.
If you use Rasa but haven’t downloaded Rasa X, you should check it out. With this release, lookup tables, regexes, and response templates more advanced than a single text response are all available in Rasa X. Previously, the absence of these features required a cumbersome workaround—now with them integrated directly in Rasa X, there will be no change to your Rasa workflow.
The new responses screen supports multiple responses per template and the option to add images and buttons, which you can preview as you edit the yaml.
Responses screen before Rasa X 0.21...
Responses screen after Rasa X 0.21...
We have also improved the UI/UX of many of the features in Rasa X:
- Conversations: inspect all of the conversations users have had with your assistant on any channel, correct any mistakes, and add these stories to your training data
- Share your bot: share a link with testers to chat with your assistant and have their interactions show up in the Conversations view
- Talk to your bot: chat with your assistant and correct its behaviour before adding it to your training data
- Stories: view, edit, and visualise stories in your training data
- Interactive Learning: provide feedback to your assistant while you talk to it and then add this story to your training data
Each screen went through extensive testing by the community, who did not shy away from sharing bugs with us, and a design review, both of which drove these recent updates. My personal favorite changes are the improved visualisations in Stories, upgrades to Talk to your Bot that let you smoothly transition to/from Interactive Learning mode, and the new channel labels in Conversations.
The new story visualisation is great because you can zoom in/out of the stories and pan over them. I am a big fan of combining stories to visualise the conversation flows represented by your stories to better understand the paths users take when chatting with your assistant.
In Talk to your Bot, you can now smoothly switch from chatting with your assistant to teaching it in Interactive Learning mode in the same conversation and easily fix any mistakes by clicking the 'Correct this' button next to any message.
The addition of a conversation channel label is a small detail, but I find it super helpful to know, especially as you both share your bot with testers and use it in production with users.
To see a list of all of the changes, check out the changelog.
What’s next for Rasa X?
The most successful product teams using Rasa follow many of the software engineering best practices routinely found in web/mobile app development. Our users tell us that one of the best elements of Rasa is the ability to use it with Git, CI/CD, etc. and so we want to make sure all of this is also possible with Rasa X.
Our future work will emphasise this: reducing the friction of integrating Rasa X into your existing workflows, enabling technical and non-technical product team members to work together smoothly. You will no longer have to choose between using Rasa X and these other tools. Instead, you will be able to use all of them.
If you want to share your workflows with us or thoughts about where we are heading, please join the discussion on the forum thread.
Check out the latest version of Rasa X for yourself today:
pip3 install rasa-x --extra-index-url https://pypi.rasa.com/simple