Run AI coding agents in isolated containers with network filtering.
Vibepit launches a sandboxed dev environment with a filtering proxy that
controls which domains your agent can reach.
[!WARNING]
Container isolation is not as strong as VM isolation. Kernel vulnerabilities,
misconfigurations, or container runtime bugs can weaken the sandbox. Vibepit
adds defense-in-depth (read-only rootfs, dropped capabilities, network
filtering) but should not be treated as a hard security boundary.
Features
- Isolated containers with read-only root filesystem and dropped capabilities
- DNS and HTTP/HTTPS filtering proxy with domain allowlist
- Network presets for common services (GitHub, Anthropic, package managers, etc.)
- Runtime allowlist management via CLI or interactive TUI monitor
- Persistent home volume across sessions
- Works with Docker and Podman
Requirements
Installation
curl -fsSL https://vibepit.dev/download.sh | bash
This downloads the latest release for your platform. Move the binary to somewhere in your PATH:
mv vibepit "$HOME/.local/bin/"
Usage
cd your/project/dir
vibepit
Start a session with additional domains or network presets:
vibepit -a example.com:443
vibepit -p github
Or run in daemon mode with SSH access:
vibepit up
vibepit connect
vibepit down
Add entries to the HTTP or DNS allowlist of a running session:
vibepit allow-http example.com:443
vibepit allow-dns internal.example.com
Open the interactive monitor to view proxy logs and manage the allowlist for a running session:
vibepit monitor
Documentation Site
Build or serve the MkDocs site locally:
make docs-install
make docs-build
make docs-serve
These targets use uv to manage Python dependencies.
make docs-build outputs the MkDocs site into site/.
Contributing
We welcome bug reports and feedback via GitHub Issues. Please open an issue to discuss proposed changes before creating a pull request.
AI Transparency
Vibepit uses AI coding agents to assist with planning, code, tests, and
documentation. We review all AI-assisted changes before merge. A change is
accepted only after tests pass and we verify the behavior. AI output can be
incorrect, so we remain responsible for all released code and documentation.
License
Apache 2.0 — see LICENSE for details.