shutapp›|

The Shutapp CLI ships in the shutapppackage. Install it with pip. That's the whole setup.

pip install shutapp
shutapp --version

Keep using Claude Code normally. Shutapp injects ANTHROPIC_BASE_URL for the child process, so Claude talks to the local proxy first.

shutapp claude

# pass Claude args through unchanged
shutapp claude -- --model claude-sonnet-4-5

Shutapp catches over 200 kinds of secrets: API keys, tokens, passwords, database URLs, private keys, cloud credentials, and Financials. It redacts the real value while leaving harmless example values in your docs alone.

# Anthropic sees:
Deploy with [ANTHROPIC_KEY_1] and card [CREDIT_CARD_1]

# Local tools receive the real values again when needed.

This is not just what you type. When Claude Code reads a file, runs a command, or uses a tool, any secret in that text is redacted the same way before it reaches the model. The same secrets are also protected when you install packages (next).

A single bad npm or pip package can run code the moment you install it. That code can read your environment variables, your .env, and your SSH and cloud keys, then send them to a server. shutapp install and shutapp pip run the install in a sandbox so it cannot.

shutapp install <pkg>        # npm install, sandboxed
shutapp pip install <pkg>    # pip install, sandboxed

Three things happen at once. Secret-shaped environment variables are removed before the install runs. The install is locked to a sandbox that cannot read your home directory, your project source, or .env. Its network is limited to the package registry. A malicious package finds no secrets and has nowhere to send them.

shutapp box status      # the sandbox in use on this machine
shutapp allow <host>    # permit one extra download host a build needs

This contains code at install time. It is not antivirus and does not scan packages, and npx, pnpm, and yarn are not sandboxed yet.

shutapp claude [--] [claude args]     # Claude Code, secrets stripped
shutapp codex [--] [codex args]       # Codex CLI, same protection
shutapp install [npm args]            # npm install inside the box
shutapp pip install <pkg>             # pip install inside the box
shutapp box status                    # the sandbox in use on this machine
shutapp allow <host>                  # permit one extra download host
shutapp session list | show | purge
shutapp detectors list [--family cloud] [--json]
shutapp status                        # includes your lifetime catch count

shutapp license activate <key>        # from your dashboard
shutapp protect [--strict]            # AUTO: everything automatic
shutapp doctor                        # AUTO: self-heal the setup
shutapp report [--since 7d]           # PRO: local "what we caught" feed

Session commands are value-free: they show ids, timestamps, placeholder counts, and detector types, never the vault contents. The catch report is the same: types and counts only, stored locally, never uploaded.

Everything runs locally on your machine. Shutapp never needs your provider key, and your secrets never leave your machine. They stay in a local session vault and are swapped for placeholders before anything reaches the model.

shutapp session list
shutapp session show <session-id>
shutapp session purge <session-id> --yes

Paste a secret into Claude Code and you'll still see it on your screen. That's your own terminal, on your machine. The model never receives it. Ask Claude to repeat it back and it can't. It only ever saw a placeholder.

you    ›  my key is sk-ant-api03-9F4xC7L2bN…
you    ›  what key did I just give you?
claude ›  I only received [ANTHROPIC_KEY_1], never the real value.