3 Top ways to build AI-based chatbots with open source software 2022

Since the early days of chatbots, bot makers have tried to develop frameworks to ease the job of creating simple and reusable components.

We’ve seen great open-source frameworks such as botkit, the microsoft bot framework and botfuel,

Some of them are still getting updates and going forward.

But as of the year 2022, the dominance has moved into the “smart”, machine learning-based, open-source framework.

Rasa (formerly Rasa NLU)

Rasa NLU is an open-source framework for chatbots, active for many years which was one of the first to provide AI-based open source capabilities for building chatbots.

It is still actively maintained with reported over 10m downloads, 600 contributors, and a great community.

Link to the code

Pros

• Great tutorials and learning materials
• Actively maintained and developed
• Customizable – offers many ways to implement your chatbot stategy, including custom NLU components

Cons

• Not fully open source – today Rasa includes tnon-open source components such as Rasa-X
• Inadequate infrastracture for management – the bots are managed by YAML files, which is not a scalable way to work with production bots

BotFront

Addressing the issues with Rasa, the team of BotFront built an open-source alternative on top of Rasa.

With all the required tooling around it, such as a UI to create examples, train and write stories.

All that with the goal to create natural dialogue flows, add rich examples and let engineers focus on systems and API integration.

Link to the code

Pros

• Beautiful UI for management
• Multi-language support
• Build your responses on the fly
• Websc

Cons

• Not actively maintained – the project was archived about a year ago
• The docs are lacking – some explainations are missing

Kairon

Another recent and interesting project, built on top of Rasa is Kairon.

Kairon is an open-source chatbot framework that aims to provide a no-coding self-service framework for creating and adapting AI assistants to specific domains.

It allows you to manage the bot through a nice UI, including stories, actions slots and even email actions and google actions.

You can see some features that are coming from recent research such as data augmentations and new bert based models and others.

Kairon is built for two personas. One can use it directly via our hosted website, and the other can host the chatbot trainer themselves using docker-compose.

You can even customize your model pretty easily with the customizable pipeline visualization

Link to the code

Pros

• Actively maintained
• Analytics module
• Great UI
• Data augmentations built in

Cons

• Relatively new – due to that, it might be a bit buggy and some things might not be developed
• No RTL support for now