webcoyote/sandvault — Run AI agents isolated in a macOS user account and sandbox-exec.

Description and use

WebCoyote SandVault is an open‑source sandbox that isolates AI coding agents—such as Claude Code, OpenAI Codex, Cursor Agent, and Google Gemini—inside a dedicated macOS user account and a sandbox‑exec profile. By leveraging macOS user separation and fine‑grained sandbox rules, it provides a secure environment for running these agents without risking the host system. The project targets developers who need a safe, reproducible way to experiment with powerful AI code assistants.

Problem

webcoyote/sandvault helps add AI capability without starting from a blank model stack.

Outcome

Early or unclear: treat as research material until maintenance, releases, docs, and issue activity are verified.

Workflow

• prototype AI features
• build RAG or agent workflows
• evaluate model tooling

Evidence

Run AI agents isolated in a macOS user account and sandbox-exec. Configured to run Claude Code, OpenAI Codex, Cursor Agent, Google Gemini.

Source