OpenShell

Active

GitHub Rust Apache-2.0

Description

OpenShell is the safe, private runtime for autonomous AI agents, developed by NVIDIA. Provides controlled execution environments and resource management.

Key Features

Sandboxed execution environments protecting data, credentials, and infrastructure with declarative YAML policies
Four defense-in-depth layers: filesystem, network, process, and inference policy enforcement
Hot-reloadable network and inference policies via `openshell policy set` without restarting sandboxes
Privacy-aware LLM routing that keeps sensitive context on sandbox compute via the Privacy Router
Provider system for managing agent credentials (API keys, tokens) injected as environment variables at runtime
GPU passthrough support for local inference and fine-tuning workloads (experimental)

Use Cases

💡 Running untrusted AI coding agents (Claude, Codex, Cursor) in isolated sandboxes to protect host systems

💡 Enforcing L7 network policies on agent outbound traffic to prevent data exfiltration

💡 Managing agent credentials securely without exposing API keys to the sandbox filesystem

💡 Deploying agent sandboxes on Kubernetes with declarative policy-as-code governance

Strengths & Limitations

✅ Strengths

• Actively maintained, recent updates
• High community interest (8.5k stars)
• Permissive open-source license (Apache-2.0)

⚠️ Limitations

• High issue backlog (548 open issues)

Quick Start

1. Install: `curl -LsSf https://raw.githubusercontent.com/NVIDIA/OpenShell/main/install.sh | sh`
2. Create a sandbox: `openshell sandbox create -- claude`
3. Connect to the sandbox: `openshell sandbox connect <sandbox-name>`
4. Apply a network policy: `openshell policy set <name> --policy policy.yaml --wait`

Visit GitHub Visit Website View Docs

Related Projects

Microsandbox

8.1k · Rust

Active A+

Secure, local, cross-platform and programmable sandboxes for AI agents. Provides strict resource isolation using microVM technology.

rustagenttools +2

· Hardware-level isolation with microVM technology
· Instant startup with average boot times under 100ms
· Embeddable SDKs for Rust, Python, TypeScript, and Go

View full details → View on GitHub

AI Agents From Scratch

4.6k · JavaScript

Normal A

Demystify AI agents by building them yourself. Local LLMs, no black boxes, real understanding of function calling, memory, and ReAct patterns.

javascriptagentevaluation +2

· 12-step progressive learning path from raw LLM calls to ReAct agents
· Runs entirely on local LLMs via node-llama-cpp, no API keys needed
· Covers function calling, memory, Tree of Thought, and error handling

View on GitHub

AgentGateway

4.7k · Rust

Active A

Next generation agentic proxy for AI agents and MCP servers. Provides unified traffic management, routing, and security control.

rustmcpagent +3

· Unified LLM gateway with OpenAI-compatible API routing to OpenAI, Anthropic, Gemini, Bedrock and more
· MCP gateway with tool federation, stdio/HTTP/SSE transports, OpenAPI integration and OAuth authentication
· A2A gateway enabling secure agent-to-agent communication with capability discovery and task collaboration

View on GitHub

Agent Governance Toolkit

6.2k · Python