By default, Gitpod uses a standard Docker Image called
Workspace-Full as the foundation for workspaces. Workspaces started based on this default image come pre-installed with Docker, Go, Java, Node.js, C/C++, Python, Ruby, Rust, PHP as well as tools such as Homebrew, Tailscale, Nginx and several more.
If this image does not include the tools you need for your project, you can provide a public Docker image or your own Dockerfile. This provides you with the flexibility to install the tools & libraries required for your project.
You can define a public Docker image in your
.gitpod.yml file with the following configuration:
The official Gitpod Docker images are hosted on Docker Hub.
You can find the source code for these images in this GitHub repository.
For public images, feel free to specify a tag, e.g.
image: node:buster if you are interested in a particular version of the Docker image.
For Gitpod images, we recommend you do not specify a tag or use
:latest to make sure you automatically benefit from security patches and fixes we release.
This option provides you with the most flexibility. Start by adding the following configuration in your
image: file: .gitpod.Dockerfile
Next, create a
.gitpod.Dockerfile file at the root of your project. The syntax is the regular
Dockerfile syntax as documented on docs.docker.com.
A good starting point for creating a custom
.gitpod.Dockerfile is the
FROM gitpod/workspace-full # Install custom tools, runtime, etc. RUN brew install fzf
Docker support: If you use the
gitpod/workspace-full image, you get Docker support built-in to your environment.
If you want a base image without the default tooling installed then use the gitpod/workspace-base image.
FROM gitpod/workspace-base # Install custom tools, runtime, etc. RUN brew install fzf
When you launch a Gitpod workspace, the local console will use the
gitpod user, so all local settings, config file, etc. should apply to
/home/gitpod or be run using
USER gitpod (we no longer recommend using
You can however use
sudo in your Dockerfile. The following example shows a typical
.gitpod.Dockerfile inheriting from
FROM gitpod/workspace-full # Install custom tools, runtime, etc. RUN sudo apt-get update \ && sudo apt-get install -y \ ... \ && sudo rm -rf /var/lib/apt/lists/* # Apply user-specific settings ENV ...
Once committed and pushed, Gitpod will automatically build this Dockerfile when (or before) new workspaces are created.
See also Gero’s blog post running through an example.
.gitpod.Dockerfile is a regular Dockerfile, you can build the image in your Gitpod workspace. This helps you catch syntax or build errors before you commit your changes.
To test your custom
.gitpod.Dockerfile, run the following commands from the project root:
docker build -f .gitpod.Dockerfile -t gitpod-dockerfile-test .
docker run -it gitpod-dockerfile-test bash
This builds a
gitpod-dockerfile-test image and starts a new container based on that image. At this point, you are connected to the Docker container that will be available as the foundation for your Gitpod workspace. You can inspect the container and make sure the necessary tools & libraries are installed.
To exit the container and return back to your Gitpod workspace, type
Once you validated the
.gitpod.Dockerfile with the approach described in the previous chapter, it is time to start a new Gitpod workspace based on that custom image.
The easiest way to try out your changes is to push them to a branch and then start another workspace on that branch, keeping the first workspace open as your main editing workspace.
Caution: The above is important in case your Dockerfile has bugs and prevents Gitpod from starting a workspace.
On start of the second workspace, the Docker build will start and show the output. If your Dockerfile has issues and the build fails or the resulting workspace does not look like you expected, you can force push changes to your config using your first, still running workspace and simply start a fresh workspace again to try them out.
We are working on allowing Docker builds directly from within workspaces, but until then this approach has been proven to be the most productive.