Give AI Agents Their Own Dedicated Computer

LLMs have reasoning capabilities, but to actually perform tasks, an independent computing environment including filesystem, shell, package manager, and more is essential. Going beyond a simple chat interface to execute code, fix errors, and form feedback loops is the core of a Production Agent.

However, giving agents a local environment or a typical Docker container is risky. Agents inherently execute untrusted code, and since containers share the kernel with the host, they are vulnerable to security threats such as Kernel Exploit.

LangSmith Sandboxes is designed to solve this problem. This solution provides not just a simple container, but a microVM-based sandbox that offers hardware-level isolation.

• Serverless-level fast startup: Instant execution environment so agents don't feel latency
• Stateful: A complete machine environment that can persist dependency installations and file modifications
• Hardware isolation: Each agent has its own independent computer with its own kernel, enhancing security