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

Using Docker

We provide a Dockerfile which allows you to build an image of Rasa Core with a simple command: docker build -t rasa_core .

The default command of the resulting container starts the Rasa Core server with the --core and --nlu options. At this stage the container does not yet contain any models, so you have to mount them from a local folder into the container’s /app/model/dialogue and app/model/nlu directories. The full run command looks like this:

docker run \
   --mount type=bind,source=<PATH_TO_DIALOGUE_MODEL_DIR>,target=/app/dialogue \
   --mount type=bind,source=<PATH_TO_NLU_MODEL_DIR>,target=/app/nlu \
   rasa_core

You also have the option to use the container to train a model with

docker run \
   --mount type=bind,source=<PATH_TO_STORIES_FILE>/stories.md,target=/app/stories.md \
   --mount type=bind,source=<PATH_TO_DOMAIN_FILE>/domain.yml,target=/app/domain.yml \
   --mount type=bind,source=<OUT_PATH>,target=/app/out \
   rasa_core train

You may in addition run any Rasa Core command inside the container with docker run rasa_core run [COMMAND].