Agent Safehouse

Active
GitHub Shell Apache-2.0

Description

Sandbox your local AI agents so they can only read and write what they need. File system permission control for secure local agent execution.

Key Features

  • Deny-first sandboxing with composable policy profiles
  • Built-in profiles for Claude, Codex, Amp and other coding agents
  • Machine-specific overrides via shell wrappers and appended profiles
  • Auto-detection of Git worktrees for cross-tree inspection
  • Symlink-aware system path resolution for macOS compatibility
  • Installable via Homebrew or standalone script

Use Cases

💡 Running AI coding agents with restricted filesystem access
💡 Enforcing least-privilege policies for LLM tool use on macOS
💡 Protecting sensitive files when using autonomous agents
💡 Shared development environments with per-developer policy customization
💡 CI/CD pipelines requiring sandboxed agent execution

Tags

sandbox security agent-tools local-ai filesystem

Categories

⚡ Agent Tools 🛡️ Security & Guardrails 🏗️ Sandbox & Execution

Quick Start

Install via Homebrew: brew install eugene1g/safehouse/agent-safehouse, then run: safehouse claude --dangerously-skip-permissions to sandbox your agent
Visit GitHub Visit Website

Related Projects