Skip to main content

How to Download and Install Cogment

Repository Latest Release Latest Docker Release

Installation methods

Installation script (compatible with linux and macOS for x86_64 architectures)

This installation method is compatible with virtually any Linux distribution, macOS and WSL2 on Windows. It only requires bash and curl.

First, download the install script and make sure you can run it using

curl --silent -L --output
chmod +x

You can then download and install the latest version to /usr/local/bin using

sudo ./

Other installation methods are available, for example:

  • Install version 2.0.0
    sudo ./ --version 2.0.0
  • Install preleased version 2.3.1-beta3, usefull because prereleases are not considered by default
    sudo ./ --version 2.0.0
  • Download the cogment exec in the current folder, make sure it is executable but skip the installation
    ./ --skip-install
  • Specify manually the arch and os
    ./ --arch amd64 --os Windows

All install options can be listed using

./ --help

Uninstall is as simple as running:

sudo rm $(which cogment)

Yes, it is possible to install Cogment as a one-line by piping the install script to bash ... at the risk of angering your system admin.

curl --silent -L | sudo ./

Manual installation (compatible with linux, macOS and Windows for x86_64 architectures)

If you are a Windows user (and also if you prefer to do a manual install), you can go through those instructions.

  1. Download the desired version from here for your platform.
  2. Copy it as cogment in a location that already belongs to your PATH (e.g. /usr/local/bin) or that you'll add to your PATH, and make sure it is executable (e.g. using chmod +x /usr/local/bin/cogment).


A Cogment docker image is available on Docker Hub. It can be retrieved with the following:

docker pull cogment/cogment

The docker version of Cogment is very handy for server deployment, however, because of the nature of docker, users need to take care of port forwarding and volume mounting if needed.

Unsupported platform

If your platform is not supported, especially if you are using an arm64 architecture, add an issue listing your platform details and do not hesitate to contact us.

Building from source

Building from sources is perfectly possible. Please refer to the build instructions for more information.

Test your installation


Depending on your platform and how you installed Cogment, it might be accessible as cogment, ./cogment, cogment.exe or docker run cogment/cogment. We will use cogment for the rest of this page and throughout the documentation.

With a working installation you can run the following in a terminal:

cogment version

You can then list all the commands by typing:

cogment help

or for help on each individual command:

cogment help <command>

Launching a Cogment app


These instructions are only compatible with unix-like environments such as Linux, macOS and WSL2 on Windows.

In order to test that your installation is fully working, run an existing Cogment app, for example one of the steps of the tutorial.

Download or clone the sources for the official Rock-Paper-Scissors (RPS) tutorial from

To run this example you will need to have the following installed on top of Cogment:

Once it is done, run the following in the directory you retrieved:

cd 5-human-player
./ build
./ services_start

The first call to ./ command will copy the cogment.yaml, and every referenced proto file, to each module directory, create virtualenvs, and install the python dependencies.

The second will start Cogment and the different services for this app. In another terminal you can connect to it and play a few games of RPS against a simple AI agent.

./ client_start

Congratulations, you have a working installation of Cogment! We recommend you head to the Cogment tutorial to learn how to implement this RPS app from scratch.


Cogment is composed of different modules, further documented in dedicated pages:

  • The orchestrator is the heart of Cogment, it orchestrates the execution of trials over actors and environments provided by ditributed services and clients.
  • The trial datastore is a storage and retrieval layer for the data generated by trials.
  • The model registry is a registry of AI models, that can be used to store, version, and make available models to AI actors.
  • The legacy cogment init, cogment copy, cogment run & cogment generate commands are still available but has been deprecated.