We’ve added full support for executing DeepAgent code inside remote sandboxes — providing safe, isolated, reproducible environments for complex agent workflows. DeepAgents can now run commands, create files, and perform long-running or potentially unsafe operations without touching your local machine.

This initial release includes native integrations with Runloop, Daytona, and Modal, with simple setup via the DeepAgents CLI (--sandbox and --sandbox-setup). Sandboxes offer:

• Safety: Isolate arbitrary or untrusted code from your local environment

• Clean, configurable environments: Spin up the exact dependencies, languages, or OS you need

• Parallel execution: Run multiple agents simultaneously without conflicts

• Long-running tasks: Offload heavy work without blocking your machine

• Reproducibility: Guarantee consistent execution across your team

Support for sandbox teardown, human-in-the-loop workflows, and secret-handling best practices are included. More configuration options and additional providers coming soon.

Read more: https://blog.langchain.com/execute-code-with-sandboxes-for-deepagents/