LangChain is releasing Deep Agents Deploy in beta - an open-source, model-agnostic way to run AI agents in production without tying yourself to a proprietary platform.
The pitch is straightforward: tools like Claude Managed Agents bundle memory and orchestration behind closed APIs, which makes switching models or platforms expensive. Deep Agents Deploy is designed to avoid that. Memory is stored in open formats (<a href="http://AGENTS.md" target="_blank" rel="noopener nofollow">AGENTS.md</a> and skills files), it works with every major model and sandbox provider, and the whole thing is MIT-licensed for both Python and TypeScript.
A single <code>deepagents deploy</code> command handles multi-tenant orchestration, sandbox setup, and exposes 30+ endpoints out of the box, including MCP, A2A, Agent Protocol, human-in-the-loop, and memory endpoints. Sandbox support covers Daytona, Modal, and Runloop. If you want to keep memory fully under your own control, self-hosting LangSmith Deployments is an option.
The core features:
<ul>
<li>Deep Agents harness</li>
<li>Choice of sandbox provider (Daytona, Modal, Runloop)</li>
<li>Short and long-term memory in open formats</li>
<li>Agents exposed via MCP and A2A</li>
</ul>
It’s currently in beta. The open standards angle is the main differentiator - if you’re already wary of lock-in with managed agent platforms, this is worth a look.

Collections

LangChain is releasing Deep Agents Deploy in beta, an open-source and model-agnostic platform for running AI agents in production. It positions itself as an alternative to proprietary solutions like Claude Managed Agents by storing memory in open formats (AGENTS.md and skills files), supporting all major models and sandbox providers (Daytona, Modal, Runloop), and using an MIT license for Python and TypeScript. A single CLI command handles multi-tenant orchestration, sandbox setup, and exposes 30+ endpoints including MCP, A2A, Agent Protocol, and human-in-the-loop. Self-hosting via LangSmith Deployments is also supported for full memory control.

LangChain launches Deep Agents Deploy as an open alternative to Claude Managed Agents