> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nemo/gym/_mcp/server.

# Infrastructure

Learn how to deploy NeMo Gym and plan cluster resources for training.

#### [Deployment Topology](/infrastructure/deployment-topology)

Server deployment patterns and training framework integration.

deployment

topology

training-integration

#### [Sandbox API](/infrastructure/sandbox)

Provider-neutral isolated execution for agents, resources servers, and benchmark harnesses.

execution

sandbox

providers

#### [Engineering Notes](/infrastructure/engineering-notes)

Technical notes on infrastructure decisions and design rationale.

engineering-notes

case-studies

## Docs

- [Deployment Topology](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/deployment-topology.md): 
- [Sandbox API](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox.md): Use the provider-neutral sandbox module for isolated execution in NeMo Gym agents and environments.
- [OpenSandbox Provider](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/opensandbox.md): Configure NeMo Gym sandboxes backed by an OpenSandbox server.
- [Apptainer Provider](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/apptainer.md): Configure NeMo Gym sandboxes backed by local Apptainer instances.
- [E2B Provider](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/e2b.md): Configure NeMo Gym sandboxes backed by E2B or an E2B-compatible gateway.
- [OpenSandbox Best Practices](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/opensandbox-best-practices.md): Resource requests vs limits, timeouts, retries, and the failure signatures that look like bugs but are not — lessons from running thousands of concurrent sandboxes.
- [Debugging OpenSandbox Sandboxes](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/opensandbox-debugging.md): How to see what CPU and memory a sandbox is getting and using, and how to detect OOM kills from the client.
- [Docker Provider](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/docker.md): Configure NeMo Gym sandboxes backed by the local Docker daemon.
- [ECS Fargate](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/ecs-fargate.md): Run sandboxes as AWS ECS Fargate tasks reached over an SSH sidecar.
- [Adding a Sandbox Provider](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/adding-a-provider.md): Implement and register a sandbox runtime backend for NeMo Gym.
- [Daytona Access](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/daytona-access.md): Obtain hosted Daytona credentials and check permissions and quota before running Gym workloads.
- [Compose Services](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/sandbox/compose.md): Run Docker Compose services through the asynchronous sandbox API.
- [Engineering Notes](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/engineering-notes.md): 
- [aiohttp vs httpx](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/engineering-notes/aiohttp-vs-httpx.md): 
- [Responses API](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/engineering-notes/responses-api-evolution.md): 
- [SWE RL Case Study](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/engineering-notes/swe-rl-case-study.md): 
- [System Design](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/engineering-notes/system-design.md): How gym env start loads configuration and launches servers, how rollout collection reaches agents through Environment Servers, and how Gym shuts down.
- [Tavily Search Access and Setup](https://docs.nvidia.com/nemo/gym/v0.7.0/infrastructure/tavily-search.md): Configure search credentials, an exclusion policy, and evaluation inputs for Tavily search.