What would be the best way to deploy RASA at scale?

I am building a bot using RASA. The performance of CORE and NLU is really great! So thank you, RASA, for that. But as I move away from developing the bot to deploying the bot to production, I can’t help but wonder that “raw deployment of RASA”(as is) might just be an excruciating slow server/service. Keeping in mind that I expect the traffic to 500,000 upwards per month, how do I effectively deploy RASA at such a scale? Furthermore, what are some gotchas or “scale issues” I need to keep in mind when doing this?

TLDR: deploy rasa at scale, how to do it, what to look out for?

Thanks, let the community benefit from the answers!!


Hey @psds01! There is a number of things you will likely want to achieve in a production deployment (stateless servers, rollbacks, reproducible deployments, multiple deployment environments, etc). We built and recently launched Rasa X Enterprise Edition that includes everything you need to run Rasa at scale in a big organization, which it sounds like you will want to look into using. @Dominik will be happy to chat with you further about this or you can fill in your information here and someone will get in contact with you

Thanks @tyd. I would love to chat further about this: scale capabilities, fast retraining of models, pricing etc. How can we do that privately?

will send you a message :+1:

1 Like