Managed Deep Agents

Managed Deep Agents is a hosted runtime for deploying and operating code-first Deep Agents in LangSmith, pairing the Deep Agents harness with managed infrastructure. It lets you run a production agent without standing up your own agent server or infrastructure. You author an agent in Python or TypeScript, then use the mda CLI to test and deploy it to the managed runtime. The managed runtime provides:

• Durable runs
• LangSmith sandboxes
• Context Hub-backed instructions, skills, and memory
• Traces
• Hosted LangGraph deployment

To deploy your first agent, see the quickstart.

​

When to use Managed Deep Agents

Choose the path that matches your control and infrastructure needs:

​

Structure your agent project

A Managed Deep Agent is a local project directory. A file’s location determines its role: the CLI reads the directory to find the agent entry, managed instructions, skills, connectors, schedules, and sandbox configuration, then packages everything into a hosted deployment. For the full directory layout and packaging rules, see the CLI project file reference. For how the CLI compiles this directory and what a deploy creates, see How Managed Deep Agents work.

​

Recommended workflow

1. Install managed-deepagents for Python or TypeScript.
2. Create a local code-first agent project with mda init.
3. Put the agent system prompt in instructions.md.
4. Add authored tools, middleware, schedules, skills, MCP connectors, and an optional sandbox.
5. Use mda dev to test your agent locally in LangSmith Studio, then mda deploy to deploy to LangSmith.
6. Inspect the deployment, traces, and runtime state in LangSmith.

New to Managed Deep Agents? Start with the quickstart, then build a complete agent step by step in the tutorial.

​

Beta notes and limits

Operational notes that apply during private beta. Behavior may change before general availability.

​

Supported models

Pass model identifiers in the form {provider}:{model_id}. For example, openai:gpt-5.5. The runtime resolves models with init_chat_model, so any provider that init_chat_model supports is usable from Managed Deep Agents, as long as the runtime has credentials for that provider. See Supported providers and models for the current list. Put local keys in .env, export them in your shell, or configure them as LangSmith workspace secrets before deploying.

​

Context Hub memory

Managed memory is stored in the same Context Hub repo as the deployed instructions and skills, at /memories/AGENTS.md. Deploy syncs instructions.md and skills/**, but preserves existing memories/** files and does not overwrite runtime-created memory. Set disableMemory: true or disable_memory=True to disable only the built-in agent-scoped memory. For how memory persists, see How Managed Deep Agents work.

​

Rate limits and quotas

During private beta, Managed Deep Agents does not publish per-key, per-workspace, or per-agent request rate limits. For workspace-specific limits, contact your LangChain team at the address in your beta access email.

​

Support and feedback

Beta access includes direct support. The contact for bug reports and feature requests is included in the email you receive when access is granted.

​

Private beta scope

Managed Deep Agents is available on LangSmith Cloud in the US region only during private beta. Self-hosted and Hybrid deployments are not supported.

​

Next steps

---