Sample chatbot with rasa stack

It uses the rasa stack (rasa core/nlu/actions) to implement a simple bot which responds user messages.

Run on local machine

Using docker

Both action server and rasa-core runs as separate processes in the same container. Rename the .env-sample to .env and optionally update the API_KEYs, TOKENs etc.

docker build -t rasa-chatbot .
docker run -it --rm -p 5005:5005 --env-file $(pwd)/.env rasa-chatbot

It starts a webserver with rest api and listens for messages at localhost:5005

Test over REST api

curl --request POST \
  --url http://localhost:5005/webhooks/rest/webhook \
  --header 'content-type: application/json' \
  --data '{
    "message": "Hi"
  }'

Response

HTTP/1.1 200 OK
Content-Type: application/json
Content-Length: 59
Access-Control-Allow-Origin: *

[{
  "recipient_id": "default",
  "text": "Hi, how is it going?"
}]

Try the bot using in terminal

docker run --rm --volume $(pwd):/app --env-file $(pwd)/.env -it\
          --workdir /app rasa-chatbot bash ./scripts/start_shell.sh

Run using docker compose

Optionally to run the actions server in separate container start the services using docker-compose. The action server runs on http://action_server:5055/webhook (docker's internal network). The rasa-core services uses the config/endpoints.local.yml to find this actions server

docker-compose up

Train Rasa model

The repository already contains a trained Rasa model at models directory. To retrain the model you can run:

docker run --rm --volume $(pwd):/app \
          --workdir /app rasa/rasa \
          train --fixed-model-name rasa-model \
          --config config.yml

Deploy to Heroku

On heroku free tier we can start two containers using two dynos, but there isn't a way for the containers to communicate with each other on Heroku. So, we push everything (actions server/rasa core/nlu) in the same container.

heroku container:push -a rasa-chatbot web
heroku container:release -a rasa-chatbot web

Another option would be to create a separate app altogether for actions server (nlu server can also be run as a separate app), which then can communicate with each other over http.

Integration with Facebook

rasa supports integration with multiple channels. Apart from exposing the REST api over http, we can integrate with facebook.

Go to https://developers.facebook.com and creat an app. We can handle messages sent to a facebook page from our app. To do so add messenger to the facebook app and subscribe to a page. Update app secret and page token in config/credentials.yml. On the facebook app, update the webhook url to the deployed heroku app (https://rasa-chatbot.herokuapp.com/webhooks/facebook/webhook).