RasaHQ/rasa-demo

Sara - the Rasa Demo Bot

:surfer: Introduction

The purpose of this repo is to showcase a contextual AI assistant built with the open source Rasa framework.

Sara is an alpha version and lives in our docs, helping developers getting started with our open source tools. It supports the following user goals:

• Understanding the Rasa framework
• Getting started with Rasa
• Answering some FAQs around Rasa
• Directing technical questions to specific documentation
• Subscribing to the Rasa newsletter
• Requesting a call with Rasa's sales team
• Handling basic chitchat

You can find planned enhancements for Sara in the Project Board

👷‍ Installation

To install Sara, please clone the repo and run:

cd rasa-demo
make install

This will install the bot and all of its requirements. Note that this bot should be used with python 3.6 or 3.7.

🤖 To run Sara:

Use rasa train to train a model (this will take a significant amount of memory to train, if you want to train it faster, try the training command with --augmentation 0).

Then, to run, first set up your action server in one terminal window:

rasa run actions --actions actions.actions

There are some custom actions that require connections to external services, specifically SubscribeNewsletterForm and SalesForm. For these to run you would need to have your own MailChimp newsletter and a Google sheet to connect to. See the development section for instructions on providing credentials for external services.

In another window, run the bot:

docker run -p 8000:8000 rasa/duckling
rasa shell --debug

Note that --debug mode will produce a lot of output meant to help you understand how the bot is working under the hood. To simply talk to the bot, you can remove this flag.

If you would like to run Sara on your website, follow the instructions here to place the chat widget on your website.

To test Sara:

After doing a rasa train, run the command:

rasa test nlu -u test/test_data.json --model models
rasa test core --stories test/test_stories.md

👩‍💻 Overview of the files

data/core/ - contains stories

data/nlu - contains NLU training data

actions - contains custom action code

domain.yml - the domain file, including bot response templates

config.yml - training configurations for the NLU pipeline and policy ensemble

Development

To install requirements for development, run:

make install-dev

To run custom actions locally, put a file called .env in the root of your local directory with values for the following environment variables. Most actions will work without them, but if you are working on actions connecting to external APIs you will need credentials.

GDRIVE_CREDENTIALS=#json access key for Google Drive API for action_submit_sales_form
MAILCHIMP_LIST=#id of mailchimp list for action_submit_subscribe_newsletter_form
MAILCHIMP_API_KEY=#api key for mailchimp
ALGOLIA_APP_ID=#algolia app ID for action_docs_search 
ALGOLIA_SEARCH_KEY=#algolia search key
ALGOLIA_DOCS_INDEX=#algolia search index
RASA_X_HOST=#Rasa X domain e.g. localhost:5002
RASA_X_PASSWORD=#password for authenticating into Rasa X
RASA_X_USERNAME=#username for authenticating into Rasa X
RASA_X_HOST_SCHEMA=#Rasa X address schema (http/https)

To run unit tests for custom actions:

make test-actions

To ensure proper database cleanup during testing, you will need to include a connection URL for your tracker store database in your .env file e.g.

TRACKER_DB_URL=postgresql:///tracker

This is not necessary for running the actions.

⚫️ Code Style

To ensure a standardized code style we use the formatter black.

If you want to automatically format your code on every commit, you can use pre-commit. Just install it via pip install pre-commit and execute pre-commit install in the root folder. This will add a hook to the repository, which reformats files on every commit.

To reformat files manuallly execute

make formatter

:gift: License

Licensed under the GNU General Public License v3. Copyright 2018 Rasa Technologies GmbH. Copy of the license. Licensees may convey the work under this license. There is no warranty for the work.