Frequently Asked Questions
Here you can find answers to a few of the most commonly asked questions about Rasa.
Where can I see examples of assistants built with Rasa, or try out a Rasa assistant?
Does Rasa support non-English languages and local dialects?
Can a Rasa assistant learn vocabulary specific to my domain or industry?
Does Rasa have any pre-built assistants I can use to get started?
Can I integrate backend services, e.g. an internal database, knowledge graph, or CRM?
What messaging channels does Rasa support?
Can I create a voice assistant using Rasa?
What skills do I need to develop with Rasa?
What is the best way to get started with Rasa?
How much training data do I need?
Does Rasa comply with data privacy standards like GDPR and HIPAA?
Can I use Rasa to build an app or service I plan to sell?
Where can I see examples of assistants built with Rasa, or try out a Rasa assistant? ¶
To see the wide variety of AI assistants that have been built with Rasa, visit the Community Showcase. The Showcase profiles both commercial and not-for-profit assistants and includes assistants in many languages and channels, including voice. Each Showcase page includes a video and a description of the assistant, and some include links you can follow to talk to the assistant or view the source code.
Rasa also maintains several chatbots in production, which are live and available for you to talk to. Sara can be found on the Rasa docs site, where she answers questions about getting started with Rasa and helps users sign up for the newsletter or contact Sales. She can also search the docs and the forum to help users find the right technical resource. Carbon bot is a project maintained by the Rasa Research team. It helps travelers calculate the carbon emissions generated by their flight and directs them to a site where they can purchase offsets.
Does Rasa support non-English languages and local dialects? ¶
Rasa assistants can be trained on any language, for any local dialect, because Rasa gives you the option to train word embeddings from scratch, using your own data. DIET, the Rasa component responsible for intent classification and entity extraction, is language-agnostic.
The Rasa NLU pipeline configuration is modular and allows for extensive customization. You can substitute tokenizers and pre-trained word embeddings specific to your language. If no pre-trained word embeddings exist for your language, you can still train the model entirely on your own data, which allows you to generate Rasa NLU models, no matter which language(s) you are working with.
Can a Rasa assistant learn vocabulary specific to my domain or industry? ¶
Because the Rasa NLU model trains on your data, your assistant can learn to recognize terms that are specific to your industry and use case, even if those terms have different meanings outside of your domain. For example, given training examples to learn from, an assistant for the insurance industry can recognize that premium refers to the cost of a policy within the context of a customer’s account.
Does Rasa have any pre-built assistants I can use to get started? ¶
Rasa offers several starter packs, which are pre-built assistants you can use as an example or a starting point for your own development. These include the Financial Services starter pack, an example banking chatbot that can help users transfer money and pay a credit card bill, and the Helpdesk starter pack, an IT helpdesk chatbot that integrates with Service Now and can open an incident report. Both of these assistants are open source and available to download.
Can I integrate backend services, e.g. an internal database, knowledge graph, or CRM? ¶
Rasa’s flexible backend can be integrated with Zendesk, Salesforce, Service Now, Hubspot, or any other third party system that provides an API. Rasa Open Source and Rasa X also include REST APIs of their own, allowing third party platforms to access the data and functionality of the Rasa stack. See the docs for more info.
The Rasa stack includes built in support for databases like MongoDB, Dynamo, and Oracle, as well as the ability to build custom database connectors. Domain data stored in knowledge graphs, also known as knowledge bases, can be accessed via built-in Knowledge Base actions.
What messaging channels does Rasa support? ¶
Rasa offers built-in channels for many common messaging platforms, including:
- Facebook Messenger
- Twilio (WhatsApp)
- Microsoft Bot Framework (Microsoft Teams)
- Cisco Webex
- Google Hangouts
If you plan to embed your chat widget on a website, Rasa supports several open source web chat options which can be customized to match the look and feel of your website.
A single Rasa assistant can be connected to multiple messaging channels at once, allowing you to serve users across different platforms without duplicating work.
Can I create a voice assistant using Rasa? ¶
Yes! Rasa Open Source integrates with text-to-speech and speech-to-text technologies, allowing you to use Rasa to process incoming messages and manage your voice assistant’s responses. You can read more about building voice assistants with Rasa here and here.
What skills do I need to develop with Rasa? How much Python do I need to know? ¶
You don’t need to be an experienced Python developer or machine learning expert to get started with Rasa. To build a simple project, you’ll need some familiarity with installing packages on the command line and writing markdown/YAML in a text editor. To build features that require custom actions (including integration with external systems), you will need to be comfortable writing Python.
Teams building production-grade bots also benefit from expertise in content creation and conversation design, as well as data science and DevOps. But for most teams getting started, some background in Python development is sufficient.
And for those with advanced programming skills and machine learning knowledge, it’s possible to completely customize Rasa Open Source, unlocking the full power of the framework.
What is the best way to get started with Rasa? ¶
You can find a collection of resources for getting started in the Rasa Documentation Portal, including installation guides, tutorials, and videos.
The Rasa Masterclass is a 12-video series covering the process of building a complex AI assistant from beginning to end. For those who prefer to follow along with a written tutorial, the Masterclass also includes an ebook.
The Building Assistants tutorial from the Rasa docs is another great resource for new Rasa developers.
We also recommend joining the Forum, where you can ask questions and learn from other developers in the Rasa community.
How much training data do I need? ¶
As a rule of thumb, you should have about 10 training examples per intent to get started.
Keep in mind that the quality of your data is more important than the quantity. It’s better to start with a small amount of data and build up your data set over time (based on conversations between your assistant and test users) than it is to autogenerate synthetic data. You can learn more about best practices for designing training data here.
Does Rasa comply with data privacy standards like GDPR and HIPAA? ¶
The entire Rasa stack can be hosted on the infrastructure of your choosing, whether on-premise or on a private cloud. This allows you to manage the full lifecycle of sensitive data, including encryption and storage. None of your data is ever sent to Rasa.
While Rasa alone is not enough to confer compliance, the ability to self-host the Rasa stack means your compliance doesn’t depend on a third-party SaaS vendor. Rasa-powered virtual assistants operate in many heavily regulated industries today, including healthcare and financial services.
Can I use Rasa to build an app or service I plan to sell? ¶
Rasa Open Source is open source software under the Apache 2.0 license, and it can be used in whole or in part to build commercial software, free of charge. Rasa X can be used for some commercial purposes, subject to restrictions. For full details, see Licensing FAQs on the Pricing Page.