Rasa chatbot with two-way integration to Robocorp digital workers

This is an example how to set your Rasa chatbot up to be able to trigger the execution of Robocorp digital workers (aka robots) through Control Room, and how the robot is able to return it's results back to the conversation. This example pairs with the robot repo.

The example chatbot can have a simple conversation and fetch the current time in difference cities around the world. Like this:

Me: Hi!
Bot: Hi, how are you?
Me: What time is it in Lisbon?
Bot: Wait a sec I'll get it. (starts a robot execution)
...
Bot: Time in Lisbon is 11:35!

Triggering the robot execution

Triggering the robot execution happens using Custom Actions. The majority of things happen in actions/actions.py file, but there are a few other things to notice.

Declare the custom action(s) in domain.yml, like this:

actions:
  - action_show_time

Also remember to make sure that you have necessary configuration for intents, stories and nlu that will trigger the action. We will not cover the setup of these in details in this example, please look for help from Rasa documentation. The code in this repo contains all you need to try out.

Once you have your robot and process set up in Robocorp Control Room, the easiest way to get the exact API call is using the API helper.

IMPORTANT: Copy your Control Room API key to the settings.py_EXAMPLE file, and rename so that the file name is just settings.py.

Also, as the robot execution might take some time depending on the types of tasks it's performing, it would be considered a best practise to let user know that the execution is started and the results will come back later.

Returning the results of the robot execution to the conversation

Setting this up requires the use of external events, which on the Rasa configuration are also actions. The documentation and reminderbot example are really good sources to get this done. We are using a rule, intent and an action like this in our example:

- rule: Bot results
  steps:
  - intent: EXTERNAL_time_result
  - action: action_tell_time

You will also need to declare the entities that the robot is returning to the conversation in domain.yml, for example like this:

entities:
  - timeresult
  - cityresult

Accordingly, the robot will then send back an HTTP POST call to trigger the given intent that has a payload like this:

{
	"name": "EXTERNAL_time_result", 
	"entities": {
		"timeresult": "12:30",
		"cityresult": "Milan"
	}
}

Running the example

To run the example chatbot, perform these actions on three different terminal tabs (provided that you have Rasa installed). Also you should have set up the corresponding Robocorp process first.

Train Rasa language model:

rasa train

Start the custom action server:

rasa run actions

Start the callback server:

python callback_server.py

Start the Rasa chatbot server:

rasa run --enable-api

Then send messages to the chatbot from command line or from the tool of your choice:

curl --request POST \
  --url http://localhost:5005/webhooks/callback/webhook \
  --header 'Content-Type: application/json' \
  --data '{"user": "some-user", "message": "Hi!"}'

The chatbot will print it's outputs to the tab where the callback server is running. Have fun chatting!