Managed Deep Agents: The Fastest Way to Launch Deep Agents for Production

Managed Deep Agents is an API-first hosted runtime for creating, running, and operating agents based on the open-source Deep Agents. Currently available in public beta, it allows programmatic management of agents within applications or internal platforms.

Deep Agents support planning, tool use, sub-agent delegation, file writing, and long-running tasks, but in production environments, execution persistence, streaming, memory, files, tools, sandboxing, and tracing must be built separately. Managed Deep Agents manages this operational layer within LangSmith, allowing developers to focus on agent behavior itself.

Key features include:

• Managed runtime: Create and modify agents and configurations via the /v1/deepagents API, and operate durable threads and streaming runs.
• Checkpoints and Human-in-the-loop: Save execution state and support workflows requiring human approval or intervention.
• Agent context and files: Maintain the AGENTS.md, skills/, subagents/, and tools.json structure, with version control in LangSmith.
• Context Hub: Store and update persistent context across executions, such as user preferences, project information, research notes, and operational procedures.
• Tools and sandbox: Configure tools via tools.json and perform tasks requiring code execution, shell commands, and file I/O in a sandbox.
• LangSmith visibility: Automatically trace all executions to inspect tool calls and debug and improve agents.

Optionally, leveraging LangSmith Engine allows analyzing agent traces to identify improvement points in prompts and code. Agents can update operational guidelines in the Context Hub based on conversations and task results obtained during actual usage.