Projects

A project groups experiments that share the same training mode. Open Experiments in the sidebar to see your project list, then open a project to work on its experiments, results, and (for Advanced Training) pipelines.

Concept

Role

Organization

Your Arena workspace (billing, members, shared data)

Project

A goal or product line. Holds experiments and, for Advanced Training, pipelines.

Experiment

One training configuration and its runs

Agent

A deployed checkpoint you can call over HTTP (created from a finished experiment)

Finding projects

Use the Experiments entry in the sidebar.

What you want

Where in the UI

List projects

Experiments sidebar (project list)

Create a project

Experiments list → New project, or Create Project when the list is empty

Open a project

Click a project row

Edit project name and settings

Experiments list → row Edit → Update Project dialog

More screen names: Finding your way.

Creating a project

On the Experiments project list, use New project (or Create Project when the list is empty). The dialog title is Create Project. Create your first project appears on Home when you have no recent projects, not on the Experiments list.

Fields:

  • Name

  • Description

  • Training switcher: Classic RL (default) or Advanced Training. Advanced Training turns on dataset and LLM-oriented experiment flows and shows the Pipelines tab. Classic RL uses gym environments only and hides Pipelines.

Button: Create Project.

The Training choice is fixed for the project. It controls which wizard you see later.

Inside a project

After you open a project, up to three tabs appear.

Experiments holds draft and configured experiments. From here you open the wizard (New experiment or a Draft row), start or resume training, stop a run, or open logs.

Results compares finished and in-flight runs. You can open logs from here too.

Pipelines appears only when the project was created with Advanced Training.

The top bar on Experiments includes New experiment, which opens Create Experiment (Name, Description, button Create Experiment).

Classic vs Advanced Training

Classic RL (Training switcher)

Advanced Training (Training switcher)

Environment step

Gym grid

Datasets and RL Environment rows

Agent step

Standard algorithm + Neural network

LLM agent UI or supervised forms for tabular datasets

Pipelines tab

Hidden

Shown

Tabular and non-tabular datasets require an Enterprise plan on the organization.