LangSmith Deployments

Deep Agents deploy launches a production‑ready, horizontally scalable server that bundles a custom Deep Agent with a LangSmith Deployment instance. The server exposes 30+ REST and WebSocket endpoints, including MCP, A2A, Agent Protocol, Human‑in‑the‑loop, and memory endpoints, enabling agents to be called as tools, orchestrated in multi‑agent systems, guarded by human approval, and queried for short‑ or long‑term memory. The deployment is open source and model‑agnostic, allowing any LLM provider (OpenAI, Anthropic, Google, etc.) and any sandbox provider (Daytona, Runloop, Modal, etc.). AGENTS.md specifies the agent’s core instructions, and Agent Skills extend it with specialized knowledge and executable actions. Unlike Claude Managed Agents, which locks agents into a proprietary ecosystem and tightly couples memory to its own infrastructure, LangSmith Deployments keep memory self‑hosted and open, offering full control over which model, sandbox, and memory backend a deployment uses. Managed Deep Agents are hosted using LangSmith Deployments· Graphiti 生成的实体摘要