Secure Agent Sandbox
Run AI coding agents in isolated containers
Prompt injection, rogue skills, and compromised dependencies can turn AI coding agents hostile. Vibepit runs agents in hardened containers where all network traffic is filtered through an allowlist. Local only: no cloud, no accounts.
Two commands, and you're in the pit
curl -fsSL https://vibepit.dev/download.sh | bash
sudo mv vibepit /usr/local/bin/Why Vibepit
Container Isolation
Each session runs in a hardened container with a read-only root filesystem and dropped capabilities.
Filtered Networking
HTTP, HTTPS, and DNS traffic is filtered through an allowlist proxy with optional network presets.
Runtime Control
Manage allowlists and inspect live traffic with CLI commands or the interactive monitor UI.
Start Here
First Sandbox
Launch your first isolated session in a project directory.
Monitor and Allowlist
Inspect proxy logs and manage live session access.
CLI Reference
Command syntax, flags, and detailed behavior for every subcommand.
AI Coding Agents
Set up Claude Code, Codex, or Copilot with the right network presets.
Architecture
How the proxy, sandbox container, and isolated network fit together.
Security Model
Understand assumptions, boundaries, and defense-in-depth controls.