LangChain launches Managed Deep Agents in public beta

Following a private testing phase, LangChain transitioned Managed Deep Agents to public beta on August 7, offering an API-first hosted runtime designed to help developers deploy and operate complex AI agents without building custom infrastructure. While building agents has become more accessible, managing them in production remains a significant hurdle. This new service, integrated directly into LangSmith, handles the operational complexities of long-running agents, including thread management, checkpointing, streaming, and observability, allowing developers to focus entirely on agent behavior.

The service operates via an API accessible under the /v1/deepagents endpoint. It preserves the existing open-source Deep Agents project structure, allowing developers to keep their agent definitions in their own repositories. The runtime supports standard configuration files such as AGENTS.md, tools.json, and directories for skills and subagents. By storing and versioning these files in LangSmith, the platform ensures that agent definitions can evolve seamlessly over time.

A key feature of the release is the Context Hub, a managed storage system that allows agents to retain and update context across multiple runs. This enables agents to learn from real-world usage and adapt to user preferences or changing operating procedures. Additionally, the runtime supports sandbox-backed execution for secure file I/O, shell commands, and code execution. Developers can also configure human-in-the-loop approvals for specific tools defined in tools.json.

For practitioners, this release eliminates the need to build and maintain custom agent servers, file storage, and feedback loops. Every run is automatically traced in LangSmith, making it easier to debug tool calls and monitor agent performance. The system is tailored for demanding workflows such as coding assistants requiring filesystem access, research agents compiling multi-session reports, and customer support triage agents that must update their own operating notes over time.