Glossary¶
Arena reuses the same words in billing, experiments, and deployments. This page defines them once so other guides do not repeat the same blurbs.
Term |
Meaning |
|---|---|
Project |
Container for experiments (and pipelines on advanced projects). Sidebar: Experiments accordion lists projects. |
Experiment |
A training configuration: resource class, environment or dataset, algorithm, hyperparameters, optional HPO on Classic RL. Has a status (see Experiment statuses). |
Run / training job |
Work started when you click Train. Billed while compute runs. |
Checkpoint |
Saved model weights at a training step. Used to resume or deploy. |
Agent |
A deployed inference endpoint (Classic RL deployment or Advanced saved model). |
Deployment |
The live inference record behind an agent. Status on the Agents table: Undeployed, Pending, Deployed, Failed. |
Resource class |
Compute SKU (CPU/GPU, region, or bring your own compute (BYOC) provider) chosen in the wizard Resources step. |
Bring your own compute (BYOC) |
Enterprise feature: train on Kubernetes you operate (self-managed, or cloud Kubernetes such as EKS, GKE, or Nebius). Profile menu: Bring your own compute. |
Training configuration |
Everything you set in the experiment wizard (resources, environment, agent, training, and HPO on Classic RL). Must be complete before Train is available. |
Classic RL project |
Default project type: gym environments and standard RL algorithms. |
OpenEnv |
The interface language-based environments implement: the rollout client drives the environment through reset/step calls. Uploads are validated against it. |
Advanced project |
Advanced flag on project: datasets, LLM algorithms, pipelines tab. |
Organization (org) |
Billing and data boundary. Users can belong to multiple orgs and switch the active one. |
Balance |
Prepaid dollars spent while training runs. New orgs start with |
ARENA_API_KEY |
Personal access token for the Arena CLI ( |