Multi‑agent systems unlock the true power of AI by enabling collaborative behaviors that drive tangible business value. At their core, agents excel through deep integrations with existing enterprise systems, connecting data and workflows in intelligent, automated ways. IBM’s watsonx Orchestrate provides the enterprise‑grade agentic platform to bring these capabilities to life, now enriched with the flexibility of LangGraph agents.

Niklas Heidloff

IBM's watsonx Orchestrate now supports an experimental feature to import LangGraph agents natively, running them in isolated environments within the platform. These imported agents can serve as top-level or collaborator agents, allowing custom system prompts and full LangGraph flexibility. The post covers four integration approaches (chat/completions endpoints, A2A, MCP/OpenAPI/Python tools, and native LangGraph import), walks through a simple hello-world LangGraph agent example, and explains how to deploy it using the Orchestrate ADK CLI with an agent.yaml spec file. Current limitations include no access to other Orchestrate artifacts, no compiled graphs, and Python-only support.

Importing LangGraph Agents in watsonx Orchestrate