_____ _ _ ____
| ____|_ _(_) |_| __ ) _____ __
| _| \ \/ / | __| _ \ / _ \ \/ /
| |___ > <| | |_| |_) | (_) > <
|_____/_/\_\_|\__|____/ \___/_/\_\
By Cloud Exit / https://cloud-exit.com
ExitBox
Multi-Agent Container Sandbox by Cloud Exit
Run AI coding assistants (Claude, Codex, OpenCode) in isolated containers with defense-in-depth security.
Getting Started
Install ExitBox and run the interactive setup wizard to configure your environment:
# Install (Linux/macOS)
curl -fsSL https://raw.githubusercontent.com/cloud-exit/exitbox/main/scripts/install.sh | sh
# Run the setup wizard
exitbox setup
The setup wizard guides you through:
- Developer role — Frontend, Backend, Fullstack, DevOps, Data Science, Mobile, Embedded, or Security
- Languages — Pre-selected based on your role, customize as needed
- Tool categories — Build tools, databases, networking, DevOps, security, and more
- Agents — Choose which AI assistants to enable (Claude, Codex, OpenCode)
The wizard generates a tailored config.yaml with your preferences. You can re-run it at any time with exitbox setup.
Features
Security
The project's security posture is rated High / Robust, employing a "Defense in Depth" strategy:
- DNS Isolation (The "Moat"): Containers cannot resolve external domain names directly. This forces all traffic through the proxy, as the container "knows" nothing of the outside internet.
- Mandatory Proxy Usage: Since direct DNS fails, tools are forced to use the configured
http_proxy. Bypassing these variables results in immediate connection failure.
- Proxy Access Control: The Squid proxy actively inspects destinations, enforcing a strict allow/deny policy (returning
403 Forbidden for blocked domains).
- Capability Restrictions:
CAP_NET_RAW and other capabilities are dropped, preventing raw socket creation and network enumeration attacks (e.g., ping is disabled).
Core Features:
- Rootless Containers: Runs without host root privileges using Podman's user namespaces
- Alpine Base Image: Minimal Alpine Linux base (~5 MB) with a managed tool list
- Supply-Chain Hardened Installs: Claude Code is installed via direct binary download with SHA-256 checksum verification against Anthropic's signed manifest — no
curl | bash
- Squid Proxy Firewall: Proxy-based destination filtering with explicit allowlist rules
- Hard Egress Isolation: Agent containers run on an internal-only network and can exit only via Squid
- No Privilege Escalation:
--security-opt=no-new-privileges:true enforced
- Capability Dropping:
--cap-drop=ALL removes all Linux capabilities
- Resource Limits: Default 8GB RAM / 4 CPUs to prevent DoS
- Secure Defaults: No automatic mounting of sensitive SSH keys or AWS credentials
Sandbox-Aware Agents
ExitBox automatically injects sandbox instructions into each agent on container start. This tells the agent it is running inside a restricted container so it won't attempt actions that can't work (e.g., running docker, podman, or managing infrastructure).
Instructions are written to each agent's native global instructions file:
| Agent |
Instructions file |
| Claude |
~/.claude/CLAUDE.md |
| Codex |
~/.codex/AGENTS.md |
| OpenCode |
~/.config/opencode/AGENTS.md |
If the file already exists (e.g., from your own global instructions), ExitBox appends the sandbox notice once. The instructions inform the agent about network restrictions, dropped capabilities, and the read-only nature of the environment so it can focus on writing and debugging code within /workspace.
Containers
- Podman-First: Optimized for Podman (rootless, daemonless) with Docker fallback
- Multi-Agent Support: Run Claude Code, OpenAI Codex, or OpenCode
- Project Isolation: Each project gets its own containerized environment
- Development Profiles: Pre-configured environments for Rust, Python, Go, Node.js, and more
- Custom Tools: Add Alpine packages to any image via
-t flag or config.yaml
Usability
- Cross-Platform: Native binaries for Linux, macOS, and Windows
- Setup Wizard: Interactive TUI that configures your environment based on your developer role
- YAML Config: Clean, readable configuration with
config.yaml and allowlist.yaml
- Config Import: All platforms use managed config (import-only). Use
exitbox import <agent> to seed host config.
- Simple Commands: Just run
exitbox claude to get started
- CLI Shorthands: All flags have single-letter aliases (
-f, -r, -t, -a, etc.)
- Session Allow-URLs: Temporarily allow extra domains with
-a — no config file edits, no restarts
Supported Agents
| Agent |
Description |
Host Requirement |
claude |
Anthropic's Claude Code CLI |
None (installed in container) |
codex |
OpenAI's Codex CLI |
None (downloaded) |
opencode |
OpenCode AI assistant |
None (binary download) |
All agents are installed inside the container. Existing host config (~/.claude, etc.) is imported once into managed storage on first run. Use exitbox import <agent> (or exitbox import all) to re-seed from host config.
Installation
Prerequisites
- Podman (recommended) or Docker — at least one is required; Podman is preferred for its rootless, daemonless design
- For Windows: Docker Desktop provides the Docker CLI that ExitBox uses
Quick Install (Linux / macOS)
curl -fsSL https://raw.githubusercontent.com/cloud-exit/exitbox/main/scripts/install.sh | sh
This downloads the latest release binary, verifies its SHA-256 checksum, and installs to ~/.local/bin/.
Linux
# Install Podman (recommended) or Docker
sudo apt update && sudo apt install -y podman # Ubuntu/Debian
# OR: install Docker - see https://docs.docker.com/engine/install/
# Install exitbox
curl -fsSL https://raw.githubusercontent.com/cloud-exit/exitbox/main/scripts/install.sh | sh
# Run the setup wizard
exitbox setup
# Run an agent
exitbox claude
macOS
# Install Podman (recommended) or Docker
brew install podman
podman machine init && podman machine start
# OR: brew install --cask docker
# Install exitbox
curl -fsSL https://raw.githubusercontent.com/cloud-exit/exitbox/main/scripts/install.sh | sh
# Run the setup wizard
exitbox setup
Windows
ExitBox runs natively on Windows with Docker Desktop.
- Install Docker Desktop for Windows
- Download the latest
exitbox-windows-amd64.exe from Releases
- Rename to
exitbox.exe and place in a directory on your PATH (e.g., C:\Users\<you>\AppData\Local\bin\)
- Run the setup wizard:
exitbox setup
Windows (WSL2)
Alternatively, use ExitBox inside WSL2 for a Linux-native experience:
# In PowerShell as Administrator
wsl --install -d Ubuntu
Then in WSL2:
sudo apt update && sudo apt install -y podman
curl -fsSL https://raw.githubusercontent.com/cloud-exit/exitbox/main/scripts/install.sh | sh
exitbox setup
Build from Source
git clone https://github.com/Cloud-Exit/exitbox.git
cd exitbox
make build # builds ./exitbox
make install # installs to ~/.local/bin/exitbox
Quick Start
# Run the setup wizard (first time)
exitbox setup
# Navigate to your project
cd /path/to/your/project
# Run an agent (builds image automatically on first run)
exitbox claude
# Or run other agents
exitbox codex
exitbox opencode
ExitBox automatically:
- Builds the container image if needed
- Imports your existing config (
~/.claude, ~/.codex, etc.) on first run
- Mounts your project directory
- Sets up the network firewall (Squid proxy)
Commands
Setup
exitbox setup # Run the interactive setup wizard (recommended first step)
Running Agents
exitbox claude [args] # Run Claude Code
exitbox codex [args] # Run Codex
exitbox opencode [args] # Run OpenCode
Management
exitbox list # List available agents and build status
exitbox enable <agent> # Enable an agent
exitbox disable <agent> # Disable an agent
exitbox rebuild <agent> # Force rebuild of agent image
exitbox rebuild all # Rebuild all enabled agents
exitbox uninstall <agent> # Remove agent images and config
exitbox aliases # Print shell aliases for ~/.bashrc
exitbox info # Show system information
Profile Management
exitbox <agent> profile list # List available profiles
exitbox <agent> profile add <name> # Add a development profile
exitbox <agent> profile remove <n> # Remove a profile
exitbox <agent> profile status # Show current profiles
Utilities
exitbox info # Show system information
exitbox logs <agent> # Show latest agent log file
exitbox clean # Clean unused container resources
exitbox clean all # Remove all exitbox images
exitbox projects # List known projects
Options
exitbox -f claude # Disable network firewall *DANGEROUS*
exitbox -r claude # Mount workspace as read-only (safety)
exitbox -v claude # Enable verbose output
exitbox -n claude # Don't pass host environment variables
exitbox -n -e MY_KEY=val claude # Only pass specific env vars
exitbox -i /tmp/foo claude # Mount /tmp/foo into /workspace/foo
exitbox -t nodejs,go claude # Add Alpine packages to image (persisted)
exitbox -a api.example.com claude # Allow extra domains for this session
exitbox -u claude # Check for and apply agent updates
All flags have long forms: -f/--no-firewall, -r/--read-only, -v/--verbose, -n/--no-env, -i/--include-dir, -t/--tools, -a/--allow-urls, -u/--update.
Available Profiles
Profiles are pre-configured development environments. The setup wizard suggests profiles based on your developer role, or you can add them manually.
| Profile |
Description |
base |
Base development tools |
build-tools |
Build toolchain helpers |
shell |
Shell and file transfer utilities |
networking |
Network diagnostics and tooling |
c |
C/C++ toolchain (gcc, make, cmake) |
node |
Node.js runtime with npm and JS tooling |
python |
Python 3 with pip |
rust |
Rust toolchain with cargo |
go |
Go runtime (arch-aware, checksum verified) |
java |
OpenJDK with Maven and Gradle |
ruby |
Ruby with bundler |
php |
PHP with composer |
database |
Database CLI clients |
devops |
Docker CLI / kubectl / helm / opentofu / kind |
web |
Web server/testing tools |
security |
Security diagnostics tools |
flutter |
Flutter SDK |
Configuration
ExitBox uses YAML configuration files stored in ~/.config/exitbox/ (Linux/macOS) or %APPDATA%\exitbox\ (Windows).
Setup Wizard
The recommended way to configure ExitBox is through the setup wizard:
exitbox setup
The wizard generates config.yaml and allowlist.yaml tailored to your developer role. It runs automatically on first use, or you can re-run it to reconfigure at any time.
config.yaml
The main configuration file controls which agents are enabled, extra packages, and default settings:
version: 1
roles:
- backend
- devops
agents:
claude:
enabled: true
codex:
enabled: false
opencode:
enabled: true
tools:
user:
- postgresql-client
- redis
settings:
auto_update: false
status_bar: true # Show "ExitBox <version> - <agent>" bar at top of terminal
default_flags:
no_firewall: false
read_only: false
no_env: false
The status_bar setting controls the thin status bar displayed at the top of the terminal while an agent is running. It is enabled by default. Set to false to disable it.
allowlist.yaml
The network allowlist is organized by category for readability:
version: 1
ai_providers:
- anthropic.com
- claude.ai
- openai.com
# ...
development:
- github.com
- npmjs.org
- pypi.org
# ...
cloud_services:
- googleapis.com
- amazonaws.com
- azure.com
custom:
- mycompany.com
Add extra Alpine packages to your container images:
-
CLI flag (persisted automatically):
exitbox -t nodejs,python3-dev claude
-
config.yaml:
tools:
user:
- nodejs
- python3-dev
The image rebuilds automatically when tools change.
Resource Limits
ExitBox enforces default resource limits to prevent runaway agents:
What Gets Mounted
ExitBox uses managed config (import-only). On first run, host config is copied into ~/.config/exitbox/<agent>/ and all mounts come from there. Host originals are never modified. Use exitbox import <agent> to re-seed from host config at any time.
| Agent |
Managed Path |
Container Path |
| Claude |
~/.config/exitbox/claude/.claude |
/home/user/.claude |
| Claude |
~/.config/exitbox/claude/.claude.json |
/home/user/.claude.json |
| Claude |
~/.config/exitbox/claude/.config |
/home/user/.config |
| Codex |
~/.config/exitbox/codex/.codex |
/home/user/.codex |
| Codex |
~/.config/exitbox/codex/.config/codex |
/home/user/.config/codex |
| OpenCode |
~/.config/exitbox/opencode/.opencode |
/home/user/.opencode |
| OpenCode |
~/.config/exitbox/opencode/.config/opencode |
/home/user/.config/opencode |
| OpenCode |
~/.config/exitbox/opencode/.local/share/opencode |
/home/user/.local/share/opencode |
| OpenCode |
~/.config/exitbox/opencode/.local/state |
/home/user/.local/state |
| OpenCode |
~/.config/exitbox/opencode/.cache/opencode |
/home/user/.cache/opencode |
Your project directory is mounted at /workspace.
When Codex is enabled, ExitBox publishes callback port 1455 on the shared exitbox-squid container and relays it to the active Codex container, so OrbStack/private-networking callback flows work reliably.
Note: SSH keys (~/.ssh) and AWS credentials (~/.aws) are NOT mounted by default for security.
Environment Variables
| Variable |
Description |
VERBOSE |
Enable verbose output |
CONTAINER_RUNTIME |
Force runtime (podman or docker) |
EXITBOX_NO_FIREWALL |
Disable firewall (true) |
EXITBOX_SQUID_DNS |
Squid DNS servers (comma/space list, default: 1.1.1.1,8.8.8.8) |
EXITBOX_SQUID_DNS_SEARCH |
Squid DNS search domains (default: . to disable inherited search suffixes) |
Architecture
Alpine Base Image
All agent images are built on Alpine Linux. The base package list is embedded in the binary and shared by all image builds.
Alpine was chosen for:
- Small image size: ~5 MB base vs ~80 MB for Debian slim
- musl libc: Matches the native binaries shipped by Claude Code, git-delta, and yq
- Consistent package manager:
apk is used everywhere — base image, profiles, and user tools
3-Layer Image Hierarchy
base image (Alpine + tools)
└── core image (agent-specific install)
└── project image (development profiles layered on)
Each layer uses label-based caching (exitbox.version, exitbox.agent.version, exitbox.tools.hash, exitbox.profiles.hash) so rebuilds are fast and incremental.
Supply-Chain Hardened Agent Installs
Claude Code is installed via direct binary download with SHA-256 checksum verification against Anthropic's signed manifest — no curl | bash. The download URL is auto-discovered from the official installer if the hardcoded endpoint ever changes. The build aborts on any checksum mismatch.
Network Firewall
ExitBox uses a Squid Proxy container to enforce strict destination allowlisting:
- Hard egress control: Agent containers run on an internal-only network with no direct internet route.
- Proxy path: Squid is dual-homed (internal + egress networks), so outbound traffic must traverse Squid.
- Allowlist: Only destinations listed in
allowlist.yaml are permitted through the proxy.
- Fail closed: Missing or empty allowlist blocks all outbound destinations.
Configuring the Allowlist
Edit ~/.config/exitbox/allowlist.yaml and add domains to the custom list:
custom:
- mycompany.com
- api.internal.example.com
Domain formats:
example.com allows example.com and its subdomains
api.example.com allows only that host scope (and deeper subdomains)
*.example.com is accepted as wildcard syntax
8.8.8.8 allows a specific IPv4 destination
2606:4700:4700::1111 allows a specific IPv6 destination
Temporary Domain Access
Allow extra domains for a single session without editing the allowlist:
exitbox -a api.example.com,cdn.example.com claude
The domains are merged into the Squid config and applied via hot-reload (squid -k reconfigure) — no proxy restart, no container restart, no connection drop. These domains do not persist across sessions.
Disabling the Firewall
exitbox --no-firewall claude # *DANGEROUS* - disables all network restrictions
Why Podman?
| Feature |
Podman |
Docker |
| Rootless by default |
Yes |
No (requires group) |
| Daemonless |
Yes |
No (requires daemon) |
| Security |
Better isolation |
Requires daemon root |
On Windows, Docker Desktop is the primary supported runtime.
Troubleshooting
Podman: "cannot find UID/GID for user"
sudo usermod --add-subuids 100000-165535 --add-subgids 100000-165535 $USER
podman system migrate
Podman on macOS: "Cannot connect to Podman"
podman machine start
Docker: Permission Denied
sudo usermod -aG docker $USER
newgrp docker
Windows: Docker Desktop not detected
Ensure Docker Desktop is running and the docker CLI is on your PATH. You can verify with:
docker info
Re-running the Setup Wizard
If your needs change or you want to reconfigure:
exitbox setup
Uninstallation
# Remove the binary
rm -f ~/.local/bin/exitbox
# Remove configuration and data
rm -rf ~/.config/exitbox ~/.cache/exitbox ~/.local/share/exitbox
# Remove container images
podman images | grep exitbox | awk '{print $3}' | xargs podman rmi -f
# OR for Docker:
docker images | grep exitbox | awk '{print $3}' | xargs docker rmi -f
On Windows, delete exitbox.exe and remove %APPDATA%\exitbox\.
License
AGPL-3.0 - see LICENSE file.
ExitBox is open-source software licensed under the GNU Affero General Public License v3.0. Commercial licensing is available from Cloud Exit for organizations that require proprietary usage terms.
Contributing
Contributions welcome via pull requests and issues.