Alquist

http://www.alquistai.com/

Alquist is dialogue manager created by Jakub Konrád and Petr Marek.

You can test current demo at https://alquistmanager.github.io/alquist-client/?e=https://alquist.herokuapp.com&bot=demo_tel

You can use HTML client https://github.com/konrajak/alquist-client for workig with Alquist.

All information about writing your bot is here https://github.com/AlquistManager/alquist/blob/master/How%20to%20write%20your%20bot.md

You can use Alquist Editor for creating your own bot https://github.com/AlquistManager/alquist-yaml-editor

Installation

You need Python 3.

We use PyYaml. You need to install it by

pip install PyYaml

We use Flask-Cors. You need to install it by

pip install -U flask-cors

We currently use WIT.ai as our nlp service so you will need to install WIT as well

pip install wit

Alternatively we are testing nlp created using MorphoDiTa tool (Mozilla Public License 2.0) and MorfFlex Czech models (Attribution-NonCommercial-ShareAlike 3.0 Unported). You will need to install the MorphoDiTa tool package.

pip install ufal.morphodita

Additional required packages can be found in requirements.txt file.

Run

To run Alquist use command

py -3 main.py

You can change Alquist's parameters. They are stored in the config.py.

port field defines the port on which the Alquist will run.

wit_token defines which Wit.ai app will you use. For our demo use NXOFXMBCIA6YAIXNNWYXJIJPC22AK35V

debug displays contents of context while bot is running, useful for debugging, values True or False

For example:

config = {"port": 5000,
      "wit_token": "NXOFXMBCIA6YAIXNNWYXJIJPC22AK35V",
      "debug": True}

API

Request

To communicate with Alquist use POST requests to url, on which the Alquist is running. POST have to contains JSON with fields:

• text Contains user's input text. You can leave it blank, if user has nothing to say (when the Alquist starts a dialogue and we need to prompt it to start for example).
• state Name of state, where we currently are. The state is "init" at the start of dialogue usually. Must contain something every time.
• context Json object representing context. There must be empty Json object at least every time.
• session ID of session used during logging. Can be empty at the start of dialogue.
• bot Name of bot.
• payload Additional information for states in the form of JSON object.

Response

Response contains JSON with fields:

• messages List of Alquist's answers. Each answer is object.

  • type Type of message. It can be text, button, iframe or slider.

  • payload Information about message. It differs according to type of message:

    • text type
      • text Text to show
    • button type
      • label Button's label
      • next_state State to transform to after button click
      • type Type of button. This field doesn't have any specific values. You can send some string and implement button according to it in client.
    • iframe type
      • url Iframe's url
      • width Iframe's width (in the percents)
      • height Iframe's width (in the pixels)
      • scrolling Will be scrollbar visible?
      • align Iframe's align
    • slider type
      • entities List of entities to save values into context
      • default_values List of default values of sliders
      • min_value Minimal value of slider
      • max_value Maximal value of slider
      • step Size of step of one slider
      • connect Should be sliders connected by color line?
      • tooltips Should be tooltips shown above the sliders?
      • tooltips_decimals Number of decimal numbers of tooltips
      • tooltips_prefix Prefix of tooltips
      • tooltips_postfix Postfix of tooltips

  • delay Delay of massage in the milliseconds.

• state Name of new state, where we get to. Use this name in the next request (state field).
• context Json object containing updated context. Use this object in the next request (context field).
• session Generated ID of session used during logging. Use this ID in the next request (session field).

Example

Request

curl -H "Content-Type: application/json; charset=\"utf-8\"" -X POST -d "{\"text\":\"INSERT_INPUT_TEXT\", \"state\":\"INSERT_STATE_NAME\", \"context\":\"INSERT_CONTEXT_OBJECT\", \"session\":\"INSERT_SESSION_ID\"}" "INSERT_ALQUIST_URL"

Specific example:

curl -H "Content-Type: application/json; charset=\"utf-8\"" -X POST -d "{\"text\":\"Hello world!\", \"state\":\"init\", \"context\":{}, \"session\":\"\"}" "http://127.0.0.1:5000/"

Docker

Alquist is prepared to be run in a Docker container. The setup process is quite straightforward.

Building an image

Dockerfile is already present in the main directory. Run docker build -t alquist . to build the Docker image for Alquist. Alternatively, you can obtain pre-built image from Docker Hub with the docker pull alquistmanager/alquist command.

Launching a container

Once the image is built, run the run-docker-container.sh bash script. There are two parameters required: First is a port you want Alquist to run on and the second is absolute path to your .yml files.