FAQ

FAQ

Does GRDL require NVIDIA OpenShell?

No. GRDL compiles to three backends: OpenShell, Docker, or standalone HTTP. The standalone backend works with any agent framework, any runtime, any language. OpenShell is one deployment option, not a dependency.

What agents does it work with?

Any agent that can make HTTP calls. Gemma 4 + Ollama, OpenClaw, Claude Code, LangChain, CrewAI, AutoGen, custom agents. See Agent Integration Guide.

How fast is it?

Sub-microsecond per evaluation. Go benchmark: ~200-500 ns/op. HTTP round-trip on localhost adds ~0.5ms.

How much memory?

~3 MB for the engine. ~6.5 MB binary.

Can an agent bypass governance?

In OpenShell/Docker: no (kernel-level enforcement). In standalone: the agent application must respect the verdict. The sidecar is advisory in standalone mode.

Why not OPA/Rego?

OPA is a general policy engine. GRDL is specifically designed for agent governance with the Seven Laws framework, multi-backend compilation, and formal verification support. They can coexist.

Can I use it commercially?

Yes. Core is Apache 2.0. Enterprise modules (future) are BSL 1.1 with free tier for up to 10 agents.

What backends are planned?

Kubernetes admission controller and a built-in grdl-runtime with Linux namespace isolation (Approach 2) are planned.