LangChain opens Managed Deep Agents to public beta on LangSmith Cloud

LangChain opened Managed Deep Agents to public beta on August 7, 2026, a hosted runtime that turns a locally built Deep Agents prototype into a deployed, durable agent with a single command.

Developers author an agent in Python or TypeScript, test it locally, then deploy it to a managed LangSmith runtime; the beta is command-line-first and limited to LangSmith Cloud in the US region.

The runtime is meant to remove the operational scaffolding that long-running agents require. It handles durable execution and retries so an agent can resume after failures, keeps persistent memory across conversations, and runs code and file operations in a sandbox. A built-in evaluation framework called Harbor tests agent behavior, and the service adds identity and authentication management, streaming progress updates, and channel integrations including Slack and GitHub.

The pitch targets the gap between a working agent demo and a production deployment, where teams typically assemble their own retries, memory stores and sandboxes. By bundling those into one runtime, LangChain is positioning LangSmith Cloud as the place agents live once they leave a laptop.

The details come from LangChain’s own announcement; the company has not disclosed pricing, and the beta’s single-region, CLI-only limits mean it is not yet a general-availability product. Its durability, memory and sandboxing claims have not been independently tested. Whether Managed Deep Agents holds up under production load and what it costs will determine if it becomes more than a convenient on-ramp.