OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Security & Sandboxing

Agent Safehouse

Agent Safehouse is a macOS command-line tool that uses `sandbox-exec` and composable, deny-first policy profiles to restrict local LLM coding agents’ access to files and integrations. It includes profiles for coding agents and app-hosted agent workflows, with rules that can be scoped to a work directory and extended or narrowed with appended policies.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Agent Safehouse — OpenViST project profile in Agent Security & Sandboxing.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when a local coding-agent workflow needs explicit macOS file-access rules, including distinct permissions for a project directory and machine-local folders.

Who it is for

  • Developers running local LLM coding agents on macOS.
  • Teams or individuals who want to limit an agent’s file and integration access while retaining project-specific permissions.

Documented uses

  • Run a coding agent with access limited to files and integrations it needs.
  • Allow an agent to work in a selected project directory while keeping particular files, such as a project `.env`, denied.
  • Add machine-specific folder permissions or policy exceptions without putting those settings in shared project configuration.
Before you choose

Requirements & limits

The project documents installation through Homebrew or a standalone shell script. It is open source under the Apache-2.0 license.

  • It is tailored to macOS and uses sandbox-exec; the project points Linux users toward Linux-native sandboxing alternatives.
  • HOME_DIR helps render home-relative rules but does not itself grant recursive home-directory reads.
  • Appended profiles load last and can add policy rules or narrow earlier defaults with deny rules.
  • Git worktree access is detected at launch for a selected worktree root; the README notes that this snapshot does not update for already-running processes.
  • The project explicitly describes the tool as a hardening layer rather than a perfect boundary against a determined attacker.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install the standalone Safehouse script

Place the standalone Safehouse script in `~/.local/bin`.

Before you start

  • macOS
  • A POSIX-compatible shell
  • Network access to download the release script

Steps

  1. Create ~/.local/bin with mkdir -p ~/.local/bin.

  2. Download the latest safehouse.sh release to ~/.local/bin/safehouse using the documented curl command.

  3. Make the file executable with chmod +x ~/.local/bin/safehouse.

Expected outcome

The standalone script is saved as `~/.local/bin/safehouse` and marked executable.

Guide limitations

  • The documented steps install the script but do not configure a shell PATH or demonstrate launching an agent.
  • Safehouse is a hardening layer, not a perfect security boundary against a determined attacker.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for AI coding assistance

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
Agent SafehouseRun a coding agent with access limited to files and integrations it needs.CLI, macOS, Self-hostedThe project documents installation through Homebrew or a standalone shell script. It is open source under the Apache-2.0 license.
Destructive Command GuardIntercept an agent's shell or Git command before it runs.CLIThe project documents prebuilt-binary installation through shell and PowerShell installers, Homebrew, and manual release archives. The shell installer is described for Linux, macOS and Windows via WSL; a separate installer is provided for native Windows.
Cc Safety NetIntercept risky shell and filesystem actions before an agent's tool call runs.CLIThe quick start requires Node.js 18 or higher and installs into coding CLIs on the current machine. The repository displays an MIT license.
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
10 October 2026
Original website title
eugene1g/agent-safehouse · Official homepage title · retained in its source language
How we review sourcesSuggest a correction

Related options

vetto — The project names it as a Linux-native sandboxing alternative for developers seeking Linux support; its README describes support for Linux and macOS.

bubblewrap — The project names it as a Linux-native sandboxing alternative for AI coding-agent workflows.