VibeKit CLI — Safety for Coding Agents
Keeps coding agents contained — isolating execution and tracking every action. Works with Claude Code, Cursor, and the rest.
npm i -g vibekit
Open Source (MIT) • View CLI Docs
🖖 VibeKit | Live
🖖 VibeKit | Sandbox: ON | Proxy: ON | Dashboard: http://localhost:3001
✱ Welcome to Claude Code!
/help for help, /status for your current setup
cwd: /Users/karpathy/Projects/vibekit/packages/clicwd: ~/vibekit/cli
Overrides (via env):
• API Base URL: http://localhost:8080• API: localhost:8080
✱ Tip: Run /install-github-app to tag @claude right from your Github issuesTip: Run /install-github-app for GitHub integration
>Try "refactor cli.js"
? for shortcuts
Keeps coding agents contained, observable, and compliant
Wraps agents locally with isolated execution and file system controls
Local sandbox
Runs agent output in isolated Docker containers — zero risk to your local setup
Observability
Complete visibility into agent operations with real-time logs, traces, and metrics
Universal agent support
Works with Claude Code, Cursor Agent, Grok CLI, Codex CLI, OpenCode, and more.
Frequently Asked Questions
Everything you need to know about VibeKit CLI
What is VibeKit CLI?
VibeKit CLI is a safe execution layer for coding agents. It runs agents like Claude Code, Gemini, Codex, Cursor, and OpenCode in isolated containers so unsafe code can't affect your local machine.
What threats does it protect against?
VibeKit prevents unsafe AI-generated code from running directly on your computer. By sandboxing agents in Docker or Podman containers, it blocks risky code from modifying files, leaking data, or harming your environment.
How fast is the protection?
Sandboxing adds minimal overhead. You can enable it per command when working with untrusted code, or set it as the default in your configuration.
Is VibeKit CLI open source?
Yes. VibeKit is open source under the MIT license. You can install it, contribute, or extend it for your workflows.
Does it only work with certain agents?
No. VibeKit provides universal agent support. It works with Claude Code, Gemini CLI, Codex CLI, Cursor Agent, Grok, OpenCode, and more — all run in the same sandboxed way.
How does the sandbox work?
When enabled, agents run inside Docker or Podman containers with controlled filesystem and resource access. This keeps processes isolated, prevents changes to your system, and ensures reproducible execution.
How do I install and get started?
Installation is simple:Verify with
npm install -g vibekit
vibekit --version
, then run an agent through VibeKit, for example:vibekit claude --sandbox docker "Generate a REST API"
Code fearlessly with any
AI agent
The safety layer your coding workflow needs
npm i -g vibekit
Open Source (MIT) • View CLI Docs