Warning: This document is for an old version of Rasa Core.

Development with Docker

This walkthrough provides a tutorial on how to set up Rasa Core, Rasa NLU, and an Action Server with Docker containers. If you have not used Rasa before it is recommended to read the Quickstart.

1. Setup

Requirements for the tutorial:

  • A text editor of your choice
  • Docker

If you are not sure whether Docker is installed on your machine execute the following command:

docker -v && docker-compose -v
# Docker version 18.06.1-ce, build e68fc7a
# docker-compose version 1.22.0, build f46880f

If Docker is installed on your machine, the command above will print the versions of docker and docker-compose. If not - please install Docker. See this instruction page for the instructions.

2. Creating a Chatbot Using Rasa Core

This section will cover the following:

  • Setup of simple chatbot
  • Training of the Rasa Core model using Docker
  • Running the chatbot using Docker

2.1 Setup

Start by creating a directory data in your project directory. Then create a file called stories.md in this directory which will contain the stories to train your chatbot:

mkdir data
touch data/stories.md

Then add some stories to data/stories.md, e.g.:

## happy_path
* greet
  - utter_greet
* mood_happy
  - utter_happy
* goodbye
  - utter_goodbye

## sad_path
* greet
  - utter_greet
* mood_unhappy
  - utter_cheer_up
* goodbye
  - utter_goodbye

After defining some training data for your chatbot, you have to define its domain. To do so create a file domain.yml in your project directory:

touch domain.yml

Then add the user intents, the actions of your chatbot, and the templates for the chatbot responses to domain.yml:

  - greet
  - mood_happy
  - mood_unhappy
  - goodbye

  - utter_greet
  - utter_happy
  - utter_cheer_up
  - utter_goodbye

    - text: "Hi, how is it going?"
    - text: "Great, carry on!"
    - text: "Don't be sad. Keep smiling!"
    - text: "Goodbye!"

2.2 Training the Rasa Core Model

Now you can train the Rasa Core model using the following command:

docker run \
  -v $(pwd):/app/project \
  -v $(pwd)/models/rasa_core:/app/models \
  rasa/rasa_core:latest \
  train \
    --domain project/domain.yml \
    --stories project/data/stories.md \
    --out models

Command Description:

  • -v $(pwd):/app/project: Mounts your project directory into the Docker container so that Rasa Core can train a model on your story data and the domain file
  • -v $(pwd)/models/rasa_core:/app/models: Mounts the directory models/rasa_core in the container which is used to store the trained Rasa Core model.
  • rasa/rasa_core:latest: Use the Rasa Core image with the tag latest
  • train: Execute the train command within the container with
    • --domain project/domain.yml: Path to your domain file from within the container
    • --stories project/data/stories.md: Path to your training stories from within the container
    • --out models: Instructs Rasa Core to store the trained model in the directory models which corresponds to your host directory models/rasa_core

This should have created a directory called models/rasa_core which contains the trained Rasa Core model.

2.3 Testing

You can now test the trained model. Keep in mind that there is currently no Rasa NLU set up. Therefore, you have to explicitly specify the user intent using the / prefix, e.g. /greet. Use the following command to run Rasa Core:

docker run \
  -it \
  -v $(pwd)/models/rasa_core:/app/models \
  rasa/rasa_core:latest \
  start \
    --core models

Command Description:

  • -it: Runs the Docker container in interactive mode so that you can interact with the console of the container
  • -v $(pwd)/models/rasa_core:/app/models: Mounts the trained Rasa Core model in the container
  • rasa/rasa_core:latest: Use the Rasa Core image with the tag latest
  • start: Executes the start command which connects to the chatbot on the command line with
    • --core models: Defines the location of the trained model which is used for the conversation.

3. Adding Natural Language Understanding (Rasa NLU)

This section will cover the following:

  • Creation of Rasa NLU training data
  • Training of the Rasa NLU model using Docker
  • Connecting Rasa Core and Rasa NLU
  • Adding a custom NLU pipeline

3.1 Adding NLU Training Data

To add some Rasa NLU training data, add a file nlu.md to your data directory:

touch data/nlu.md

Then add some examples to each intent, e.g.:

## intent:greet
- hey
- hello
- hi
- good morning
- good evening
- hey there

## intent:mood_happy
- perfect
- very good
- great
- amazing
- wonderful
- I am feeling very good
- I am great
- I'm good

## intent:mood_unhappy
- sad
- very sad
- unhappy
- bad
- very bad
- awful
- terrible
- not very good
- extremely sad
- so sad

## intent:goodbye
- bye
- goodbye
- see you around
- see you later

3.1 Training the NLU Model

You can then train the Rasa NLU model by executing the command below. As output of the command the directory models/rasa_nlu will contain the trained Rasa NLU model:

docker run \
  -v $(pwd):/app/project \
  -v $(pwd)/models/rasa_nlu:/app/models \
  rasa/rasa_nlu:latest-spacy \
  run \
    python -m rasa_nlu.train \
    -c config.yml \
    -d project/data/nlu.md \
    -o models \
    --project current

Command Description:

  • -v $(pwd):/app/project: Mounts your project directory into the Docker container so that the chatbot can be trained on your NLU data.
  • -v $(pwd)/models/rasa_nlu:/app/models: Mounts the directory models/rasa_nlu in the container which is used to store the trained Rasa NLU model.
  • rasa/rasa_nlu:latest-spacy: Using the latest Rasa NLU which uses the spaCy pipeline .
  • run: Entrypoint parameter to run any command within the NLU container
  • python -m rasa_nlu.train: Starts the NLU training with
    • -c config.yml: Uses the default NLU pipeline configuration which is provided by the Docker image
    • -d project/data/nlu.md: Path to the NLU training data
    • -o models: The directory which is used to store the NLU models
    • --project current: The project name to use.

3.2 Connecting Rasa Core and Rasa NLU

You can connect Rasa Core and Rasa NLU by running each container individually. However, this setup can get quite complicated as soon as more components are added. Therefore, it is suggested to use docker compose which uses a so called compose file to specify all components and their configuration. This makes it possible to start all components using a single command.

Start with creating the compose file:

touch docker-compose.yml

The file starts with the version of the Docker Compose specification that you want to use, e.g.:

version: '3.0'

Each container is declared as a service within the docker compose file. The first service is the rasa_core service.

    image: rasa/rasa_core:latest
      - 5005:5005
      - ./models/rasa_core:/app/models
      - start
      - --core
      - models
      - -c
      - rest

The command is similar to the docker run command in section 2.4. Note the use of the port mapping and the additional parameters -c rest. The ports part defines a port mapping between the container and your host system. In this case it makes 5005 of the rasa_core service available on port 5005 of your host. This is the port of the REST Channels interface of Rasa Core.

The parameters -c rest instruct Rasa Core to use REST as input / output channel. Since Docker Compose starts a set of Docker containers it is not longer possible to directly connect to one single container after executing the run command.

Then add the Rasa NLU service to your docker compose file:

    image: rasa/rasa_nlu:latest-spacy
      - ./models/rasa_nlu:/app/models
      - start
      - --path
      - models

This maps the Rasa NLU model in the container and instructs Rasa NLU to run the server for the model.

To instruct Rasa Core to connect to the Rasa NLU server and which NLU model it should use, it is required to create a file config/endpoints.yml which contains the URL Rasa Core should connect to:

mkdir config
touch config/endpoints.yml

Docker containers which are started using Docker Compose are using the same network. Hence, each service can access other services by their service name. Therefore, you can use rasa_nlu as host in config/endpoints.yml:

  url: http://rasa_nlu:5000

To make the endpoint configuration available to Rasa Core, you need to mount the config directory into the Rasa Core container. Then instruct Rasa Core to use the endpoints configuration with the parameter --endpoints <path to endpoints.yml> and define the targeted Rasa NLU model with -u <nlu model to use>. By adding this additional configuration to your docker-compose.yml it should have the following content:

version: '3.0'

    image: rasa/rasa_core:latest
      - 5005:5005
      - ./models/rasa_core:/app/models
      - ./config:/app/config
      - start
      - --core
      - models
      - -c
      - rest
      - --endpoints
      - config/endpoints.yml
      - -u
      - current/
    image: rasa/rasa_nlu:latest-spacy
      - ./models/rasa_nlu:/app/models
      - start
      - --path
      - models

3.3 Running Rasa Core and Rasa NLU

To start Rasa Core and Rasa NLU execute:

docker-compose up


Add the flag -d if you want to run it detached.

The REST API of Rasa Core is then available on http://localhost:5005. To send messages to your chatbot:

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

Your chatbot should then answer something like:

    "recipient_id": "default",
    "text": "Hi, how is it going?"

If the chatbot cannot understand you, the answer is [].

3.4 Adding a Custom NLU Pipeline

If you want to configure the components of your NLU Pipeline, start by creating a file nlu_config.yml in your config directory:

touch config/nlu_config.yml

Put the description of your custom pipeline in there, e.g.:

- name: "nlp_spacy"
- name: "tokenizer_spacy"
- name: "intent_entity_featurizer_regex"
- name: "intent_featurizer_spacy"
- name: "ner_crf"
- name: "intent_classifier_sklearn"

Then retrain your NLU model. In contrast to the previous training also mount the config directory which contains the NLU configuration and specify it in the run command:

docker run \
  -v $(pwd):/app/project \
  -v $(pwd)/models/rasa_nlu:/app/models \
  -v $(pwd)/config:/app/config \
  rasa/rasa_nlu:latest-spacy \
  run \
    python -m rasa_nlu.train \
    -c config/nlu_config.yml \
    -d project/data/nlu.md \
    -o models \
    --project current

Then adapt the NLU start command in your docker compose so that it uses your NLU configuration. As in for the training mount the config directory into your NLU container and instruct Rasa NLU to use this configuration by adding the flag -c <path to your nlu config>. The configuration of the rasa_nlu server might then look similar to this:

    image: rasa/rasa_nlu:latest-spacy
      - ./models/rasa_nlu:/app/models
      - ./config:/app/config
      - start
      - --path
      - models
      - -c
      - config/nlu_config.yml

Depending on the selected NLU Pipeline you might have to use a different Rasa NLU image:

  • rasa/rasa_nlu:latest-spacy: To use the spaCy pipeline
  • rasa/rasa_nlu:latest-tensorflow: To use the tensorflow_embedding pipeline
  • rasa/rasa_nlu:latest-mitie: To use a pipeline which includes mitie
  • rasa/rasa_nlu:latest-full: To build a pipeline with dependencies to spaCy and TensorFlow
  • rasa/rasa_nlu:latest-bare: To start with minimal dependencies so that you can then add your own

4. Adding Custom Actions

To create more sophisticated chatbots you will probably use Actions. Continuing the example from above you might want to add an action which tells the user a joke to cheer the user up.

4.1 Creating a Custom Action

Start with creating the custom actions in a directory actions:

mkdir actions
# Rasa Core SDK expects a python module.
# Therefore, make sure that you have this file in the direcotry.
touch __init__.py
touch actions/actions.py

Then build a custom action using the Rasa Core SDK, e.g.:

import requests
import json
from rasa_core_sdk import Action

class ActionJoke(Action):
  def name(self):
    return "action_joke"

  def run(self, dispatcher, tracker, domain):
    request = requests.get('http://api.icndb.com/jokes/random').json() #make an api call
    joke = request['value']['joke'] #extract a joke from returned json response
    dispatcher.utter_message(joke) #send the message back to the user
    return []

Next add the custom action in your stories and your domain file. Continuing the example from above replace utter_cheer_up in data/stories.md with the custom action action_joke and add action_joke to the actions in the domain file.

4.2 Adding the Action Server

The custom actions are run by the action server. To spin it up together with Rasa Core and Rasa NLU, add a service action_server to the docker-compose.yml:

  image: rasa/rasa_core_sdk:latest
    - ./actions:/app/actions

This pulls the image for the Rasa Core SDK which includes the action server, mounts your custom actions into it, and starts the server.

As for Rasa NLU, it is necessary to tell Rasa Core the location of the action server. Add this to your config/endpoints.yml:

  url: http://action_server:5055/webhook

Run docker-compose up to start the action server together with Rasa Core and Rasa NLU and to execute your custom actions.

4.3 Adding Custom Dependencies

If your action has additional dependencies, either systems or python libraries, you can add these by extending the official image.

To do so create a Dockerfile, extend the official image and add your custom dependencies, e.g.:

# Extend the offical Rasa Core SDK image
FROM rasa/rasa_core_sdk:latest

# Add a custom system library (e.g. git)
RUN apt-get update && \
    apt-get install -y git

# Add a custom python library (e.g. jupyter)
RUN pip install --no-cache-dir \

You can then build the image and use it in your docker-compose.yml:

docker build . -t <name of your custom image>:<tag of your custom image>

5. Adding a Custom Tracker Store

By default all conversations are saved in-memory. This mean that all conversations are lost as soon as you restart Rasa Core. If you want to persist your conversations, you can use different Tracker Stores.

5.1 Using MongoDB as Tracker Store

Start by adding MongoDB to your docker-compose file. The following example adds the MongoDB as well as a UI (you can skip this), which will be available at localhost:8081. Username and password for the MongoDB instance are specified as rasa and example. For example:

  image: mongo
  image: mongo-express
    - 8081:8081

Then add the MongoDB to the tracker_store section of your endpoints configuration config/endpoints.yml:

  store_type: mongod
  url: mongodb://mongo:27017
  username: rasa
  password: example

Then start all components with docker-compose up.

5.2 Using Redis as Tracker Store

Start by adding Redis to your docker-compose file:

  image: redis:latest

Then add Redis to the tracker_store section of your endpoint configuration config/endpoints.yml:

  store_type: redis
  url: redis

5.3 Using a Custom Tracker Store Implementation

If you have a custom implementation of a tracker store you have two options to add this store to Rasa Core:

  • extending the Rasa Core image
  • mounting it as volume

Then add the required configuration to your endpoint configuration config/endpoints.yml as it is described in Tracker Stores. If you want the tracker store component (e.g. a certain database) to be part of your docker compose file, add a corresponding service and configuration there.

Have questions or feedback?

We have a very active support community on Rasa Community Forum that is happy to help you with your questions. If you have any feedback for us or a specific suggestion for improving the docs, feel free to share it by creating an issue on Rasa Core GitHub repository.