LangChain releases Deep Agents deploy in beta

LangChain has released Deep Agents deploy in beta, enabling developers to publish open-source AI agent harnesses to production using a single command line prompt. Built on an MIT-licensed framework for Python and TypeScript, the system is designed as an open alternative to proprietary offerings like Claude Managed Agents. The framework packages agent orchestration logic and multi-tenant scaling through a single command, requiring configuration parameters such as AGENTS.md for instructions, mcp.json for tools, alongside specialized skill definitions and sandbox preferences.

The framework operates independently of any single AI provider. Developers can connect their agents to models from OpenAI, Google, Anthropic, Azure, Bedrock, Fireworks, Baseten, Open Router, or Ollama. Execution sessions can run inside isolated sandboxes provided by Daytona, Runloop, Modal, or LangSmith Sandboxes, among other third-party services. Under the hood, each deployment automatically spins up a production-ready LangSmith Deployment server that exposes 30+ API endpoints. These endpoints support open protocols including MCP over HTTPS/SSE, A2A for multi-agent setups, Agent Protocol for building user interfaces, human-in-the-loop guardrails, and direct memory querying.

The primary advantage of this open structure centers on context retention and preventing vendor lock-in. While switching between model providers is relatively easy, proprietary agent platforms often trap an agent's memory and learning flywheels inside closed ecosystems. By utilizing standard formats like AGENTS.md and allowing teams to self-host their LangSmith Deployments, LangChain ensures that crucial interaction data remains stored entirely within a developer's own database infrastructure.