Managed Deep Agents

Overview

Managed Deep Agents is a hosted runtime for deploying and operating code-first Deep Agents in LangSmith. You author an agent in Python or TypeScript, then use the mda CLI to test it locally and deploy it to the managed runtime. It pairs the open-source Deep Agents harness (see [[deep-agents-core]]) with managed infrastructure: durable runs, LangSmith sandboxes, Context Hub-backed instructions, skills, memory, traces, and hosted LangGraph deployment.

The core idea is that an agent is a directory. A file's location determines its role, and the CLI compiles that directory into a managed LangGraph app. There is no API-driven create/update/invoke flow during private beta: you write code and run mda deploy.

When to use

Use this skill when the user wants to:

• Build a Deep Agent in code (Python or TypeScript) and deploy it to LangSmith without standing up their own server.
• Add authored tools, middleware, MCP connectors, cron schedules, skills, or a sandbox to a managed agent.
• Test an agent locally with mda dev and deploy it with mda deploy.
• Understand what the managed runtime owns versus what the author configures.

Use a standard LangSmith Deployment (see [[langgraph-cli]], langgraph deploy) instead when the user needs custom application code, custom routes, advanced authentication, stronger isolation, maximum scalability, or a region other than US LangSmith Cloud.