Managed Deep Agents: the fastest way to ship a production deep agent
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
LangChain is launching Managed Deep Agents, a hosted runtime for running production-grade deep agents within LangSmith. It handles the operational complexity of long-running agents — durable execution, checkpointing, streaming, sandboxed tool access, human-in-the-loop workflows, and observability — so developers can focus on agent behavior rather than infrastructure. Agents are defined using the open-source Deep Agents harness (AGENTS.md, skills/, subagents/, tools.json) and deployed via an API-first interface at /v1/deepagents. A Context Hub feature lets agents persist and update working context across runs, enabling continuous improvement from real usage. LangSmith Engine can review traces to find bugs and update agent context automatically. Use cases include support triage, research, coding, data analysis, and internal ops agents. Now available in public beta.
Table of contents
Why Managed Deep AgentsWhat we're launching in private betaHow it worksWhat this unlocksBuilt on open-source Deep AgentsGet startedQuestions this post answers
What infrastructure does LangChain Managed Deep Agents handle so I don't have to build it myself?
Managed Deep Agents handles durable threads, checkpointing, streaming runs, human-in-the-loop workflows, sandbox-backed execution, tool configuration, and LangSmith observability. Developers keep the agent definition (AGENTS.md, skills/, subagents/, tools.json) in their own repo and interact with the runtime via the /v1/deepagents API, without deploying a custom agent server. Teams productionizing AI agents track managed runtime options like this on daily.dev.
What is Context Hub in LangChain Managed Deep Agents and how does it help agents improve over time?
Context Hub is a managed storage layer within Managed Deep Agents where agents persist and update working context across runs — things like user preferences, project details, research notes, or operating procedures. Combined with LangSmith Engine, which reviews agent traces to find bugs and improvement areas, agents can update their own Context Hub files between runs based on real usage patterns. Developers building self-improving agents follow developments like Context Hub on daily.dev.