The `nvidia/openshell-community` repository serves as a community-driven ecosystem built around NVIDIA's OpenShell, a runtime environment designed for autonomous AI agents. OpenShell itself is a secure and private platform where these agents can operate, learn, and iterate. This community repository extends the core functionality of OpenShell by providing a hub for contributed resources, integrations, and pre-built environments.

The primary purpose of OpenShell is to provide a safe and controlled environment for autonomous agents to function. It acts as a programmable factory, enabling agents to generate synthetic data, test edge cases, and safely navigate failures within isolated sandboxes. The core engine of OpenShell, which is housed in a separate repository, includes the sandbox runtime, a policy engine, a gateway (with a k3s harness), a privacy router, and a command-line interface (CLI). This community repository complements the core engine by offering additional tools and resources to enhance the agent development and deployment process.

The main features of this community repository revolve around providing pre-built components and integrations to streamline the use of OpenShell. It includes launchables, such as a Brev integration for one-click cloud deployment, allowing users to quickly set up and run OpenShell instances. The repository also offers pre-built sandbox images tailored for specific domain-specific workloads. These sandboxes provide ready-to-use environments with pre-installed tools and skills, accelerating the development and testing of autonomous agents.

The repository currently features several pre-built sandboxes, including a foundational image with essential system tools, a sandbox for integrating with Ollama (a tool for running local and cloud LLMs), a sandbox for synthetic data generation workflows, and a sandbox for open agent manipulation and control (OpenClaw). These sandboxes offer a variety of starting points for developers to experiment with different agent functionalities and integrate them within the OpenShell environment.

Getting started with OpenShell Community requires a few prerequisites, including the OpenShell CLI, a compatible container runtime like Docker, and an NVIDIA GPU with appropriate drivers for GPU-accelerated images. The repository provides a quick start guide using Brev, a platform that simplifies the deployment process. Users can deploy OpenShell on Brev instances with either CPU-only or NVIDIA H100 GPUs, depending on their needs. The Brev integration allows for remote OpenShell gateway deployments, external inference endpoints, and all-in-one playgrounds for sandboxes and UI workflows.

The repository also provides instructions on how to create and use the pre-built sandboxes. Users can create a sandbox using the `openshell sandbox create` command, specifying the desired sandbox image from the available options. For example, the Ollama sandbox allows users to run local LLMs and route to cloud models, with Claude Code and Codex pre-installed. The repository also encourages community contributions, with a dedicated CONTRIBUTING.md file outlining the guidelines for contributing to the project. Security is also a priority, with a SECURITY.md file detailing the process for reporting vulnerabilities. The project is licensed under the Apache 2.0 License.