What’s New?

March 2019

Rasa Platform

We held back some changes from February’s release to make sure they were just right, so this one’s a big one.

There’s been a massive UX/UI overhaul! Although some parts of it are still underway, we’ve combed over every square pixel of the platform and tried to make a design that is consistent, discoverable, easy-to-use and (of course) beautiful 🌇. We hope you like it!

We now support multiple deployment environments, so you can move NLU and Core models from staging to development to production, testing them at each step of the way.

Annotating your data just got much easier. See a message that needs more time to fix than you currently have? You can flag it for yourself and others to see and address later. But expect to have a little more time because you can now change annotations right from the conversation window - just click a user message and make your corrections in the panel on the right. And what could be a more perfect pairing with these new features than an annotator user role, so you can let someone loose on tidying up your data without letting them train models or change the model you’re using in production.

Note that this is a beta release of our next major version and so relies on Core version 0.14.0a2 and NLU version 0.15.0a4.

As always, if you want more details there are technical changelogs over in the Platform docs.

February 2019

New developer content

While we are working on improvements on Rasa NLU and Core, we have published a few pieces of developer content. Make sure to check them out and learn more about:

January 2019

Released Rasa Core 0.13:

Released Rasa NLU 0.14

  • ability to save successful predictions and classification results to a JSON file from rasa_nlu.evaluate
  • analyser parameter to intent_featurizer_count_vectors featurizer to configure whether to use word or character n-grams.

From now on, Rasa Core only supports Python 3. Rasa NLU will support Python 2.7 up until version 0.14.

December 2018

Rasa and Botsociety integration!

New integration between Rasa and Botsociety designed to bridge the gap between the conversational AI designers and developers. Using this new integration you can design the conversation and test the conversation on Botsociety, and seamlessly transition to the development stage with Rasa. Learn more about this integration and see it all in action by watching the recording of our livestream.

Rasa Platform improvements!

Conversation analytics landed, so you can track your results over time. You’ll find a link to the page at the top of the sidepanel.

Notice something in a conversation that you need to address at some point in the future? Now you can hover over a line of a conversation and flag it for later. And to help you find your flagged messages and much more, we introduced conversation filtering. Just expand the filter at the top of your conversations page to narrow down the chats by intent, policy, action and more.

Aside from those big additions, we’ve been working on speed and reliability improvements as always. We also now fully support emoji in your Training Data so your 🦸🏿 can 🥯 without your 🧶 getting 🚡.

November 2018

Major release of Rasa Core (0.12) which brings a lot of new functionality:

  • Vastly improved slot filling through the new Forms. You can now separate out business logic from learned behavior.
  • A new interactive learning experience which renders a dynamically updated graph to the web browser.
  • End-to-end evaluation of Rasa NLU and Core models together.

New interactive learning feature in Rasa Platform. Jump from any point in a user conversation into interactive learning, providing corrections to the bot’s actions and taking over as the user.

October 2018

Improved support for multiprocessing in the Rasa NLU server. Tensorflow training is now non-blocking in python 3.

All Rasa Platform users now have access to experimental features. Features can be activated via the Rasa Platform UI or programatically using the API. Experimental support for interactive learning in the Rasa Platform UI.

September 2018

The interactive learning CLI has been re-worked completely and made much more user friendly. Users and now undo annotation mistakes, make NLU corrections for both intents and entities, and navigate system predictions using arrow keys. Added a socket.io input channel to Rasa Core for real-time messaging. Rasa Core’s evaluate script can now be made to fail on prediction errors for use in CI. Added support for JWT authentication in the Rasa Core server, improving security. The UserMessage class now tracks which channel the user was active in (e.g. Facebook, Slack)

A major release (0.16.0) or Rasa Platform with full support for Rasa Core 0.11.4+ and Rasa NLU 0.13.3+. Added documentation for running Rasa Platform with custom Core and NLU components / classes. Added an API endpoint for fetching all service logs for remote debugging of Rasa Platform installations.

August 2018

Rasa Core 0.11 is out!

This is a major update, which requires a little migration work, but greatly simplifies deploying Rasa Core in production. Custom actions are now run in a separate server, which you can create using the new rasa_core_sdk.

  • Check out the detailed changelog, migration instructions, and livestream.
  • Rasa Core 0.11 also ships with the new EmbeddingPolicy, implemented in tensorflow and described in a paper (link coming soon).
  • For Rasa Platform Users: Rasa Platform 0.16 (coming September 2018) will depend on Rasa Core 0.11. Please ensure that you migrate before updating your installation. This update is necessary to enable new platform features, and to make deployment and maintenance simpler. If you have any questions about migrating to Rasa Core 0.11, please use your dedicated support email to reach out.
  • Rasa Platform can now show every conversations in story format, just click on a message in a conversation and click ‘view story’. This is a convenient way to collect more stories as training data.
  • Rasa Platform has an updated CMS for editing bot responses. Editing templates is instantaneous and doesn’t require re-training Rasa Core.

July 2018

All new docs!

  • Interactive quickstarts for Rasa NLU and Rasa Core
  • Unified search across all docs, and all docs now at https://rasa.com/docs
  • Reorganized Rasa NLU and Core docs, splitting ‘learning’ from ‘reference’ material

Community Forum

We felt that a lot of good discussions were getting lost on gitter. So we’re moving the community support over to the newly created Rasa Community Forum It’s going to be a great place for the Rasa community to meet, find answers to questions, participate in discussions, share ideas and keep up to date with the most recent Rasa updates.

Released Rasa Core 0.10

  • Ability to manage bot responses externally (instead of putting them into the domain.yml file)
  • An option to ignore entities for certain intents
  • New default action ActionDefaultFallback

Released Rasa NLU 0.13

  • language-agnostic NER
  • support multuple training processes per project