OpenShell: Secure Runtime for AI Agents

OpenShell is a secure and private runtime designed for autonomous AI agents. It offers sandboxed execution environments that safeguard your data, credentials, and infrastructure through the use of declarative YAML policies. These policies prevent unauthorized file access, data exfiltration, and uncontrolled network activity.

Key features include:

• Sandboxed Environments - Isolates each agent in its own container with policy-enforced egress routing.
• Policy Engine - Enforces constraints on filesystem, network, and process levels.
• Multi-Agent Support - Ships with built-in skills for various agents, including Claude and Codex.
• GPU Support - Experimental feature for GPU passthrough to enhance local inference capabilities.
• Real-Time Terminal UI - Provides a live dashboard for monitoring gateways and sandboxes.

OpenShell is intended for developers and organizations looking to create secure environments for AI workloads, with a focus on privacy and control over data access.