Hi all,
I have a rasa open source bot already deployed on a server. I want to use rasa X as an interface for talking to the bot and collect stories from users. That is, using only the two top options in the menu bar.
I don’t want rasa x to train the model, to load nlu data, etc. (these are too many to be handled by rasa anyway).
Is there away to do this? Rasa x is installed on the same machine, through pip.
You have a few options here:
You can connect a server-mode Rasa X installation (so not through pip, but using one of the other installation methods) to your open source deployment so that conversations flow in to Rasa X, even though the model is not being served by Rasa X. Then you can save these as stories. There are instructions for connecting an existing deployment here:
If you just want to use Rasa X for local development and not to collect production user conversations, you can run Rasa X in local mode and only use Interactive learning and share your bot. If you didn’t want to load in your whole bot data-wise, you could start it in a dummy repo that contains the necessary structure, but very little data, and then upload a model you had trained elsewhere. If you’re wanting to save conversations to your training data though, I’d imagine it makes sense to have your training data available in Rasa X, even if you don’t use it to train a model - I’m not sure what you mean re.
these are too many to be handled by rasa anyway
If you’d like to clarify that might change things.
If you don’t want to use the other features in Rasa X that’s entirely up to you, the buttons will still be there but you can just ignore them.
I finally gave a simple solution, I created a new project and copied the files (nlu, stories, rules, actions, domain). So no a direct connection to the “production” bot, but a copy (both running on the same machine).
Regarding the clarification, my bot has a somewhat big FAQ, as in ~10.000 questions (and several answers per question). My nlu.yml has ~40.000 lines, and domain.yml is ~6.000 lines. This poses challenges to the UI (too many items to anybody see/manage), and very long loading times for rasa X (~15 minutes on an amd 5950X cpu).
@mloubser Is there a way for rasa x to get the events from rasa open source? I followed the instructions and added RabbitMQ in the bot, the bot sends messages to the queue, but I am not sure how rasax can get them. I am running rasax locally, and does seem to ignore completely rabbitmq. I tried added an event broker, and I got an error that the sqlite broker cannot be changed.
Rasa X in local mode does not support streaming events from an open source deployment. You’ll need to use a server deployment method (but you can use e.g. multipass to do this on your laptop) if you want to set it up this way. The RabbitMQ referred to in the instructions is the Rasa X RabbitMQ instance, not the Rasa Open Source instance. I also got this the wrong way round the first time I tried to export events from OSS to X, so I empathize =)
Well, I patched rasa x to also get events from rabbitmq. 