Skip to content
driifPublic

About

AI Multiplexer - solves one problem: creating a secure space where AI can run wild and do whatever it wants with dangerous permissions.

Resources

Security policy

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

zwinger

zwinger runs AI coding agents (Claude Code, Codex, pi, Gemini CLI) in cages: small Linux microVMs on your Mac. Each cage belongs to a project, a folder on your Mac. An agent in a cage sees only the folders that you mount and reaches only the hosts that you allow. Your subscription logins stay on the Mac: the cage's proxy adds them to requests to the provider, and the agent sees a placeholder only.

Install

Needs a Mac with Apple silicon and macOS 14 or newer.

  1. Download the .dmg from the latest release.
  2. Drag zwinger to Applications.
  3. The app is not signed with an Apple Developer ID yet. Open it the first time with a right-click on zwinger.app, then Open. On macOS 15 and later, open it once, then select Open Anyway in System Settings > Privacy & Security. Or run xattr -dr com.apple.quarantine /Applications/zwinger.app.

zwinger installs its VM runtime itself and updates itself. It needs no Homebrew, Docker or ~/.microsandbox.

First steps

The first start shows four steps:

  1. zwinger checks the Mac (macOS version, virtualization, free disk space) and installs the VM runtime.
  2. Create your first project: pick the project folder. The agent works in it as /workspace.
  3. Add an agent: zwinger finds the agents that are installed on the Mac and adds one with your login.
  4. Start the cage and open the agent with ⌘P, or a shell with ⌘T.

⌘K shows all actions. ⌘, opens Settings: agents on this Mac, agent runtimes, updates, storage.

Screenshots

  • First run: system checks
  • Project overview with the map
  • Agent terminal
  • Network activity with a blocked host
  • Task run room

Security model

  • Each cage is a Linux microVM with its own kernel (microsandbox, libkrun on Hypervisor.framework). See ADR 0001.
  • Network: deny all, an allow list, or the public internet, per project. The app shows the hosts that the agent looks up and the ones that the cage blocks.
  • Secrets and host logins are never in the VM. The proxy injects them for their hosts only. See ADR 0002.
  • The app window has a strict Content Security Policy and minimal permissions. See docs/security.md.

Build from source

Needs Rust (stable), Node.js 24 and pnpm.

pnpm install
pnpm tauri dev

The gates:

pnpm build
cargo fmt --manifest-path src-tauri/Cargo.toml --check
cargo clippy --manifest-path src-tauri/Cargo.toml -- -D warnings
cargo test --manifest-path src-tauri/Cargo.toml

How to make a release: docs/release.md.

License

Apache 2.0

About

AI Multiplexer - solves one problem: creating a secure space where AI can run wild and do whatever it wants with dangerous permissions.

Resources

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages