I’m building a no-code chatbot builder with a drag-and-drop editor. The interface allows users to visually build conversational flows using nodes like:
Authentication
Payment
Messages
Forms
Custom scripts
Agent handoff, etc.
Once the user builds a flow in the UI, we plan to generate a working chatbot in the background — ideally powered by a framework like Rasa Open Source.
Before going deeper, I wanted to ask:
Is Rasa Open Source a good fit for this use case?
Has anyone used Rasa to programmatically generate bots from a visual builder?
Are there known limitations or best practices when using Rasa in this kind of dynamic, schema-driven setup?
Our goal is to let non-technical users create production-ready bots without writing code, while still having a powerful, extensible backend under the hood.
Appreciate any advice or experience you can share!
I’ve used Rasa a bit, and it can definitely work for what you’re trying to do, especially since it’s open source and quite flexible. The tricky part is that turning a visual flow into something Rasa understands means you’ll need to build a solid way to convert those drag-and-drop elements into intents, stories, and training data. Also, keep in mind that retraining the model every time a user changes something can slow things down, so you might want to be smart about when and how often that happens. From my experience, error handling and fallback logic can get complicated with no-code users, so having good defaults and safety nets is key to avoid the bot breaking. Overall, Rasa is a solid choice if you’re comfortable building some backend logic to glue everything together, but it’s not exactly out-of-the-box for a fully dynamic, no-code builder.
Thanks so much for your response. For current project We’ve decided to move forward with using Rasa primarily as an intent mapper, rather than relying on it for full end-to-end dialogue management. That approach seems to give us the structure and extensibility we need, without the complexity of constantly regenerating full stories or retraining the model for every UI change.
This setup should cover most of our use cases, and we can layer in custom logic for things like fallbacks and dynamic responses outside Rasa where needed.