agent-safehouse
Sandbox your local AI agents so they can read/write only what they need
What is agent-safehouse?
Sandbox your local AI agents so they can read/write only what they need
Overview of agent-safehouse
What it is
Agent Safehouse is a macOS tool that sandboxes LLM coding agents using sandbox-exec with composable policy profiles and a deny-first model. It restricts agent access to only the files and integrations they need, based on a least-privilege philosophy. It is designed as a hardening layer, not a perfect security boundary against a determined attacker.
Key features
- Deny-all starting point, with explicit allow rules for required paths.
- Composable policy profiles for major coding agents and app-hosted agent workflows.
- Built-in system path resolution that follows symlinks for compatibility paths.
- Auto-detection of Git worktrees at launch for shared metadata access.
- Support for machine-specific overrides via appended profiles and environment variables.
- Default metadata-only traversal on root and home directories, with explicit exceptions for XDG locations and agent instruction folders.
Use cases
- Sandboxing local AI coding agents (e.g., Claude, Codex, Amp) on macOS.
- Restricting agent file access to a specific work directory while keeping system files functional.
- Sharing project-level sandbox configurations across a team while allowing per-machine overrides.
- Reducing the risk of accidental or malicious file modifications by agents.
Getting started
Install via Homebrew:bashbrew install eugene1g/safehouse/agent-safehouseOr as a standalone script:bashmkdir -p ~/.local/bincurl -fsSL https://github.com/eugene1g/agent-safehouse/releases/latest/download/safehouse.sh -o ~/.local/bin/safehousechmod +x ~/.local/bin/safehouse
License and pricing
Licensed under Apache-2.0. It is open source and free to use.
Based on the project's public README and repository metadata.