LangChain opens Managed Deep Agents in public beta on LangSmith Cloud

An agent as a project folder

According to LangChain, a Managed Deep Agent is a normal Python project. The only required file is agent.py, which defines the agent with define_deep_agent. The system prompt sits in instructions.md, and each folder under skills/ holds a playbook the agent loads when a task calls for it. Tools are ordinary Python functions under tools/, while tools/mcp.py adds tools from remote MCP servers.

Developers test the agent locally with mda dev, which starts a local Agent Server and opens the agent in LangSmith Studio to chat with it and inspect traces. The command mda deploy ships the project to LangSmith and adds the model provider's key to the deployment. The company says the product first ran on May 13, 2026, in a private beta, and the post was updated on November 2, 2026 to describe the public beta.

Production features as files

LangChain argues that building a capable agent keeps getting easier, but running it in production remains a lot of work. Agents need threads that survive restarts, a safe place to run code, memory that lasts between conversations and a record of what they did. Scheduled runs also require a cron server. Managed Deep Agents is meant to take that layer off the team's plate, and every deployment runs on LangSmith Agent Server.

Each managed capability is a file or folder, and mda deploy sets up whatever it finds:

• Memory: by default an agent remembers only the current thread. Adding memory.py keeps knowledge across threads, in a shared layer, a private layer per user, or both.
• Schedules: a file under schedules/ runs the agent on a cron schedule.
• Channels: channels/slack.py lets people start runs from Slack. Any service that can send a webhook can start runs over HTTP.
• Sandbox: sandbox/ gives the agent an isolated filesystem and shell for running the code it writes.
• Identity and connections: identity.py authenticates callers, and connections store API keys and run OAuth so end users authorize their own accounts.
• Subagents and evals: subagents are declared in the agent definition, and mda evals init -i sets up a Harbor eval workspace under evals/.

Some capabilities need no file. The agent can use built-in web search powered by Parallel without a separate account or API key, and it can work in a browser, handing control to the user when a site needs a login. Selected tool calls can pause for human approval via interrupt_on.

Calling agents and improving them

Every deployment serves the Agent Server API, so an application can create threads and stream runs, or a team can build a custom frontend. Deployments also serve an MCP endpoint at /mcp, which lets MCP clients such as Claude Code call the agent as a tool.

Instructions and skills are kept in LangSmith Context Hub. Each deploy syncs instructions.md and skills/ from the project, and they can be edited in the Hub without redeploying. If the Hub copy changed since the last deploy, the CLI asks which version to keep. Durable memory also lives in Context Hub. Runs are traced in LangSmith, and LangSmith Engine uses those traces to surface recurring issues and diagnose their cause. Connected to the agent's Context Hub repository, it can propose fixes to instructions and skills.

What teams are building and how it relates to open source

LangChain says public beta teams run incident triage agents that investigate Slack alerts in a sandbox, engineering agents started by issue tracker webhooks, internal assistants such as HR or finance helpers, customer-facing agents with per-user memory, long-running project agents and scheduled agents modeled on company functions like sales or product. The company also says its own GTM Agent, Open SWE and Content Studio run on the service, and that Content Studio's code shrank by about 80% after the move.

The managed service runs the same open-source harness, with the same agent loop, filesystem tools, subagents and skills. The difference is who compiles the agent: create_deep_agent returns a compiled agent the developer runs and hosts, while define_deep_agent returns a definition that the managed runtime compiles. Teams wanting full control over hosting can still use the Deep Agents SDK on their own infrastructure.

For developers, the pitch is a shorter path from a prototype to a deployed agent without operating servers, schedulers or memory stores. LangChain's quickstart takes users from an empty directory to a deployed research assistant, and a tutorial then adds a custom search tool, durable memory and a daily schedule.