Bots in production

Have you built a chatbot with Rasa that is now running in production? We would love to know more about it! Share with us what have you built, what were the biggest challenges when deploying it in production, what feedback do you get from the users of your bot, and anything else you would like to share.

Your feedback and experience is what helps us improve Rasa Stack and we believe that your projects will inspire the community to build great chatbots! :slight_smile:

1 Like

We built a chatbot using Rasa. It is in Alpha Testing mode. We are addressing FAQ data of our website. This is my first project in the field of AI,

I had to work hard while training the bot (data entry in story, domain files). The most important thing to take care is Training Data . I believe if training data is good, bot will answer as expected. Rasa NLU has been very helpful due to it’s pipeline feature that provides flexibility to integrate any NLP library.



Awesome! Is it currently running in production? I would love to try it out if there is a chance! :slight_smile: Also, did you use both Rasa NLU and Rasa Core for this project?

Thanks. It uses both Rasa NLU (v 0.12.3) + Core (v 0.9.8). It is live on Slack. Due to NDA with client I may not disclose other details, but YES Rasa has reduced efforts by 40% to build the chat bot. We are quite close to meet client’s expectation (working with optimisations).

Hello! I am also developing a bot that is primarily FAQ based for my company. I have developed a basic model on my system and it works well. Now the company wants to move to production phase and since this is my first project, i don’t have much idea about production and would like to know a few things: 1- What are the server requirements in terms of RAM, OS and Memory for a bot that has to handle around 100 users at a time when it is live. 2- What about scaling? Will a single instance of my bot be able to handle all the queries or do i have to use things like docker or wsgi(my front end is in Flask).


We also use RASA NLU and its components and have several (~70) live bots in production for many different clients. We are currently running into scaling issues since the traffic is getting quiet high.

1 Like

I doubt RASA stack is matured enough to build a production ready chat-bot … I tried migrating from our dialogue flow based chat bot to rasa and faced many hurdles … probably may be I am relatively new to RASA … But I feel that RASA need too much training data to at-least recognize the entities properly …The export from dialog-flow was not enough to migrate and need many modifications to at least to set up basic functionalities to work. and migrations is not seamless :-)… Even some times I see the entities in different user context getting mixed up … Eg: in your weatherbot location supplied by one user taking when a second user ask the same question !!!. Un sure if that was a bug with some specific version of rasa

1 Like

It is a difficult matter to compare Rasa with Dialogflow as with Dialogflow you are probably using many pre-built models to serve your specific task. With Rasa, you will definitely need a bit more understanding of basic chatbot concepts and how to train your models which again demystifies how chatbots can work and how you can control the outcome top-down which in some industries is extremely crucial where risk and liability is higher. Meaning we need to properly explain the inner workings of a chatbot and justify it before taking it to production. For some companies, that may not be a requirement.

I think with the Rasa stack, your team needs a bit more maturity than compared to Dialogflow. I have a bot in production of my own in Dialogflow which works absolutely fine given the need, however I cannot justify the same with my current job given the risk and liability is higher and we need to concentrate a lot more on how exactly are we able to control the outcome.

This is purely based on your needs and the outcome you are looking for in the end :slight_smile: So the question of production readiness is purely subjective to your organisational strategy.


Companies least likely allow sensitive data or financial data to handle with chat bots even it is hosted on primise. The bots which developed for such companies mostly just like a call center executive answering to users. When some thing come related to critical data mostly bot will handover the chat to actual human.These kind of use cases challenges are least. But if there are requirements which need processing of the collected data , invoking actions from rasa to an external application may not be a good option as it will add more point to point connection and failure handling will be difficult when we add more connection .Not sure if rasa provide any other better way to handle such scenarios.In google dialogue flow each response will have all the data values previously collected for that context and session.

In our organisation, all data coming from our client is considered sensitive until qualified and qualifying raw text is really difficult. You can’t really tell your clients not to write sensitive information in a free text field such as chatbots. It doesn’t work. Also, there is always a big question about processing of your data. Google do not play well in such areas.

Like i said it needs the team to mature a bit more to understand how Rasa works. We took some time but for us both cases turn by turn data collection and getting to the right response does work with Rasa. We will be in production soon.

In our case, Rasa has provided us with a better transparency on exactly how our models are performing and upto to which point we can be accurate. We get to evaluate our models something i am not able to do with dialogflow or other chatbot cases. It really boils down to how much control you would like on your chatbot stack for your use cases. For some cases I do agree i don’t need to know that much in depth about how the model is performing since the business case is not that sensitive.

p.s - i am in a financial industry :slight_smile:


Hey Vikas!

Sorry to hear about that. I’d like to hear a little more about your problems so I can forward them to the team. For the entity recognition, did you try any of the other components in your pipeline listed here?

As for the Dialogflow export, what went wrong exactly? We’re always looking for improvements

@Juste I had created chat bot on local and want to test it on production. Could you please suggest some documentation about to hosting bot on production. I am very new on development and it will be very helpful for me. Thank in advance

1 Like

Hello Juste,

I just created the mood bot and everything is working very well. I have also integrate my working bot into telegram and its working, am able to chat with my bot from the telegram app on my phone.:grinning::grin:…But the problem is that when the user mood is Sad, it doesnt retrieve the image selected by the user rather bot utters “Did that help?”''this only happens in Telegram""

Am using ngrok and port 5004. I think am getting something wrong some, i also tried running the action server along site with ngrok, but my bot wont still retrieve the image.

I tried changing the url in endpoints.yml but wont show .

Please helptelegram_bot


I am stuck at initial level.

When I m running with credentials.yml I m stuck at rasa_core server is up and running. May I know after that what are steps to follow, how to open our bot at telegram.

Same here, Lets connect and try to solve the problems.

I am actually working on a assigment to develop a bot for FAQ’s. I am stuck in some places where the main communication integration needs to be done.

@agrima_agarwal Hey, have you created the bot on telegram and retrieved the credential details following the guide here?

@adityap31 Feel free to open a question with the issues you are facing as well so we can help you out :slight_smile:

I would like to know the same things. Did you find answers to your questions? (1- What are the server requirements in terms of RAM, OS and Memory for a bot that has to handle around 100 users at a time when it is live. 2- What about scaling? Will a single instance of my bot be able to handle all the queries or do i have to use things like docker or wsgi(my front end is in Flask).)


For model evaluation, we’ve found very helpful

1 Like

We have started doing a poc on rasa stack in our company .
Our intention is to handle customers through various channels like voice,mail,sms using Rasa as a backend. We found it can solve our problem . So I gave a demo to my manager 2 days back. He is happy with rasa but he asked me the below questions , Can someone help me to find the answers -

  1. Is it possible to implement sentiment analysis ? If yes then how ?
  2. How many requests can be handle by Rasa core and NLU at a time ?
  3. Is it comparable to Google Dialogflow ?

Thanks in advance.