LLM Sandbox for Gemini CLI

io.github.vndee/llm-sandbox

Securely run LLM-generated code in isolated containers across 7 languages and 3 container backends.

client:Gemini CLI transport:stdio runtime:pypi

Install LLM Sandbox in Gemini CLI

~/.gemini/settings.json · windows: %USERPROFILE%\.gemini\settings.json

BACKEND

DOCKER_HOST

KUBECONFIG

NAMESPACE

COMMIT_CONTAINER

KEEP_TEMPLATE

SANDBOX_NETWORK_MODE

SANDBOX_READ_ONLY

SANDBOX_CAP_DROP

SANDBOX_SECURITY_OPT

SANDBOX_MEMORY

SANDBOX_CPUS

Held in this page only — never stored, logged, or sent anywhere but back to your screen.

{
  "mcpServers": {
    "llm-sandbox": {
      "command": "uvx",
      "args": [
        "llm-sandbox"
      ],
      "env": {
        "BACKEND": "<BACKEND>",
        "DOCKER_HOST": "<DOCKER_HOST>",
        "KUBECONFIG": "<KUBECONFIG>",
        "NAMESPACE": "<NAMESPACE>",
        "COMMIT_CONTAINER": "<COMMIT_CONTAINER>",
        "KEEP_TEMPLATE": "<KEEP_TEMPLATE>",
        "SANDBOX_NETWORK_MODE": "<SANDBOX_NETWORK_MODE>",
        "SANDBOX_READ_ONLY": "<SANDBOX_READ_ONLY>",
        "SANDBOX_CAP_DROP": "<SANDBOX_CAP_DROP>",
        "SANDBOX_SECURITY_OPT": "<SANDBOX_SECURITY_OPT>",
        "SANDBOX_MEMORY": "<SANDBOX_MEMORY>",
        "SANDBOX_CPUS": "<SANDBOX_CPUS>"
      }
    }
  }
}

Gemini expands $VAR and ${VAR} inside env values, so a secret can live in your shell rather than this file. Gemini CLI docs

LLM Sandbox in other clients

LLM Sandbox MCP server