Nixcage – Sandbox AI coding agents per project with Nix and direnv
direnv auto-activation with bubblewrap sandboxing isolates agents per project.
Sandbox any macOS app — only your project directory stays accessible
Native macOS sandboxing stops AI agents from reading your SSH keys without Docker overhead.
macOS developers using AI coding agents
Sandboxie · Docker Desktop · OrbStack
I built it because Docker sandboxing requires config and planning. Build in sandboxing of AI tools instead is limited to the very tools themselves, instead I wanted to have a simple cage around Claude running inside VSCode. Also needed to protect files inside a folder like .env.local or keys.
Install via: brew install holtwick/tap/bx
Run like: bx claude .
direnv auto-activation with bubblewrap sandboxing isolates agents per project.
Firecracker MicroVMs sandbox agent actions where Copilot Workspace trusts subprocess calls.
Clever launchd scheduling beats cron for maximizing AI token rate limits.
Shows all four Claude limits (Session, Weekly, Weekly Sonnet, Overage) in one widget with color-coded warnings and reset countdowns — exactly the tiny UX gap heavy users kept bumping into. The author went native: Swift + WidgetKit, OAuth PKCE (no API keys), App Group UserDefaults for widget/app sync and a Homebrew tap for install — practical choices that make this something you'd actually run on your desktop. Limited to macOS Sonoma and Claude subscribers, so great for the audience but narrow in reach.
Runs agents in sandboxed microVMs to contain execution; promising but docs barely explain why this matters.
Two-stage brake saves your repo state before Claude Code cuts you off.