NVIDIA OpenShell
- Price
- Free
- API
- No
- Weights
- Yes
WHAT IT DOES
OpenShell is NVIDIA's open-source runtime for running AI agents inside a sandbox. It sits beneath the agent harness, runs the agent without privileges, limits which files it can reach, filters system calls at the kernel level, and checks every network request against policy by binary, destination, method and path. Nothing is allowed by default, credentials are only supplied where policy permits, a policy-aware router controls which model endpoints the agent can call, and every decision is logged. It supports Claude Code, Codex, GitHub Copilot CLI, Hermes, LangChain Deep Agents, OpenClaw and OpenCode, and runs on Docker, Podman, Kubernetes via Helm, and NVIDIA DGX Spark and DGX Station. First announced in March 2026, it became the software half of NVIDIA's Open Agent Safety Platform on September 28, 2026, alongside Sentry, a monitor that runs on BlueField-4 hardware.
WHY WE PICKED IT
After a run of agents escaping their test environments, the useful question is where the fence lives. OpenShell enforces policy outside the agent's own process, where a prompt cannot talk it out of the rules, and it works with the coding agents people already use. It is free, open source under Apache 2.0, and needs no special hardware.
QUICK FACTS
STRENGTHS
- ✓Deny-by-default policy for files, network, processes and credentials
- ✓Enforcement runs outside the agent, so it cannot be prompted away
- ✓Works with Claude Code, Codex, Copilot CLI, OpenClaw and other harnesses
- ✓Apache 2.0 open source, runs on Docker, Podman or Kubernetes
- ✓Every allow and deny decision is auditable
LIMITATIONS
- ⚠Policies take real work to write well; too loose and the sandbox protects little
- ⚠Sentry, the hardware monitor in NVIDIA's full platform, needs BlueField-4 hardware most teams do not have
- ⚠Does not judge whether an allowed action is a good idea, only whether policy permits it
GETTING STARTED
Clone github.com/NVIDIA/OpenShell and follow the docs to run your agent in a sandbox under Docker or Podman. Start from a tight policy and open access only as the agent needs it; the GitHub push tutorial in the docs is a good first example.
PRICING
FEATURED IN GUIDES
RAG, Done Properly: Why Chunks Are the Wrong Unit of Knowledge
Chunk-based RAG retrieves text windows, not answers. On a 298-page benchmark, switching to question-...
Inference Engines Compared: llama.cpp, MLX, ExLlama, vLLM, SGLang, TensorRT-LLM
Inference engines are memory-bandwidth-bound schedulers, not just model loaders. Comparing llama.cpp...
How to Run an AI Agent Locally With Zero Token Costs
Perplexity and Nvidia just shipped a version of Perplexity's Computer agent that runs locally by def...
AI Infrastructure: Run, Host, or Rent?
Choosing AI infrastructure means trading control, cost, and latency against each other. Learn when t...
AI Infrastructure: Foundations at Scale
AI infrastructure is moving from "run models wherever" to specialized foundation layers. Understand ...
EXPLORE ALTERNATIVES
Compare NVIDIA OpenShell with 5+ similar infrastructure AI tools.
FREQUENTLY ASKED QUESTIONS
Is NVIDIA OpenShell free?
Yes, NVIDIA OpenShell is completely free to use.
Does NVIDIA OpenShell have an API?
No, NVIDIA OpenShell does not appear to offer a public API.
What is NVIDIA OpenShell best for?
NVIDIA OpenShell is best for Best Open-Source Agent Sandbox. OpenShell is NVIDIA's open-source runtime for running AI agents inside a sandbox. After a run of agents escaping their test environments, the useful question is where the fence lives. OpenShell enforces policy outside the agent's own process, where a prompt cannot talk it out of the rules, and it works with the coding agents people already use. It is free, open source under Apache 2.0, and needs no special hardware.
What platforms does NVIDIA OpenShell support?
NVIDIA OpenShell supports local.
Is NVIDIA OpenShell open source?
Yes, NVIDIA OpenShell is open source. You can access the source code on GitHub at https://github.com/NVIDIA/OpenShell.
How do I get started with NVIDIA OpenShell?
Clone github.com/NVIDIA/OpenShell and follow the docs to run your agent in a sandbox under Docker or Podman. Start from a tight policy and open access only as the agent needs it; the GitHub push tutorial in the docs is a good first example.
FEATURED ON CURATEDAI
Work on NVIDIA OpenShell? You're hand-reviewed in our directory. Add this badge to your site — it links back to this profile.