nolabs-ai/nono

secure multiplexed execution paths for agents - zero trust, zero setup, zero latency.

What it solves

It provides a secure, zero-latency sandbox for running AI agents (such as Claude Code, Codex, and CoPilot) without the overhead of virtual machines, containers, or daemons. It prevents agents from accessing sensitive system data—like SSH keys and cloud credentials—while allowing them to perform necessary tasks in a restricted environment.

How it works

nono uses a least-privilege sandbox that isolates the agent's execution. It goes beyond simple isolation by placing the tools the agent calls (e.g., git, gh, curl) into their own isolated child sandboxes. This means a tool can have a separate policy, filesystem grants, and network rules from the agent itself. It uses a composable JSON policy system to define exactly what an agent and its delegated tools can access, including L7 filtering for API calls and a credential proxy to inject secrets securely.

Who it’s for

It is designed for developers and engineers who run AI agents locally or in production and need to ensure their credentials and production systems remain safe without sacrificing developer velocity.

Highlights

  • Zero Setup: Runs in seconds with no VM, container, or daemon required.
  • Granular Tool Isolation: Isolates delegated tools in separate child sandboxes to prevent blanket permission grants.
  • Credential Management: Uses a credential proxy to inject secrets into tools without exposing them to the agent.
  • Cross-Platform: Supports macOS, Linux, and Windows (WSL2).
  • Registry Support: Includes a registry of pre-configured agent profiles for popular AI agents.

Related

  • Project
    finbarr/yolobox

    A containerized sandbox for AI coding agents that allows them to run commands without permission prompts while protecting the host home directory from accidental destruction.

  • Project
    akitaonrails/ai-jail

    A secure OS sandbox for AI coding agents that restricts filesystem, network, and system access to prevent host system compromise.

  • Project
    opensandbox-group/OpenSandbox

    OpenSandbox is an open‑source sandbox platform for AI agents and other code‑execution workloads. It offers multi‑language SDKs, a CLI, and a server that can run sandboxes on Docker locally or on Kubernetes at scale, with strong isolation (gVisor, Kata, Firecracker), network policies, and a credential vault. Use cases include coding agents, browser/desktop automation, RL training, and secure code‑execution services.

  • Project
    nikvdp/cco

    cco is a sandboxing wrapper for AI coding agents (Claude Code, Codex, etc.) that runs them in an isolated environment to protect your system from prompt injections and accidental damage, while preserving a seamless terminal experience.

  • Dispatch
    Securing AI Agents with Isolated Docker Sandboxes

    An exploration of agent-sandbox, a tool designed to run AI coding agents within isolated Docker containers to prevent unauthorized system access and ensure execution safety.