Environments

The agent RL environments training can reference, and what each one declares.

Every environment you may reference: public ones plus your own. Selecting one shows its manifest.

Agent environments and what each manifest declares.

What you see

Field	What it tells you
Protocol	How a trainer reaches it — served here, generated for NeMo Gym, or a remote service somebody else runs
Verified by	A rubric, a plugin entrypoint, or an external service
Harness	The plugin that drives the model through it, where the protocol needs one
Runs at	For a remote environment, the host it answers on
Sandbox	Whether it needs one, which decides whether a deployment can run it at all
Versions	Every immutable version, and which are public

The taskset is never shown, and no route returns it. What a business considers hard is often more revealing than its training data, so a job references an environment and receives task content only inside the approved runtime.

What you can do

• Reference one in a job — copy its <owner>/<name>@<version> and pass --environment.
• Publish a version — sf env push, see authoring environments.
• Register a running service — sf env runner register, for an OpenEnv service you already operate.

The protocol field, in one line each

Protocol	Who runs the server
OpenEnv	The platform. Turn-level; TRL, SkyRL, Unsloth and Axolotl consume it natively
NeMo Gym	The training framework. The platform writes the files Gym expects and Gym launches its own
Remote OpenEnv	Somebody else, already. The platform only decides whether the job may reach it
OpenEnv container	Nobody here — this deployment does not host other people’s containers, and a job requesting one is refused at admission