Running code execution in an AI agent is harder than it looks.
Deploy LangSmith Sandboxes to give agents isolated compute.
Deploy LangSmith Sandboxes to give agents isolated compute.
Summary
Agents require a real computer with filesystem, shell, package manager, and persistent state, but granting them access to infrastructure is risky. LangSmith Sandboxes provide isolated virtual computers for agents, allowing them to run code, access files, and make authenticated API calls safely. The infrastructure shift means each agent gets its own compute environment, preventing dangerous access to shared resources. Sandboxes answer how to give agents compute safely, highlighting the need for secure sandboxing when agents run millions of tasks. The solution involves containerized environments that isolate agents while providing persistent state and a controlled interface to external services.
By using LangSmith Sandboxes, developers can give agents the compute power they need without exposing the underlying infrastructure to potential abuse or data leaks.
Key changes
- Agents need isolated virtual computers for filesystem, shell, and API calls
- LangSmith Sandboxes provide safe compute environments
- Each agent receives its own compute to prevent infrastructure abuse
- Sandboxing enables millions of tasks without exposing infrastructure
- Containerized environments isolate agents while offering persistent state
- Secure sandboxing is essential for agent execution
- Sandboxes answer how to give agents compute safely
- Infrastructure shift to isolated compute per agent