Use the Agent Development Kit to build from scratch or import your existing LangGraph agents directly into watsonx Orchestrate
IBM watsonx Orchestrate gives development teams several paths to build agents and bring them to production. Agents can be created in low-code or pro-code using the Agent Development Kit (ADK). Such agents will leverage a rich set of reasoning options, such as ReAct, Flows, Collaborators, and more. Developers can also import their existing LangGraph agents directly, and immediately benefit from the enterprise-grade governance, security and observability needed to run them at scale.
Implement and call tools in your agents that enable AI agents to invoke external APIs and services, thereby extending their capabilities.
Create workflows that pause for human validation, capture context and resume automatically with audit-ready traceability.
Utilize a comprehensive evaluation framework to effectively validate agent behavior, identify regressions, and ensure stability and reliability throughout lifecycle.
Establish an automated pipeline that monitors, evaluates and retrains agents to improve accuracy and effectiveness by integrating tools like OpenTelemetry and Langfuse.
Use the AI Gateway to integrate multi-provider AI models into watsonx Orchestrate, with configurable routing policies that maximize flexibility and performance.
Build agents that interact with users through various channels and modalities, including voice and text.
Leverage the no-code AI Agent Builder to create AI agents with guided steps and templates, or build custom agents using code or visual prototyping in Langflow.
Use Orchestrate Agent Architect Prompt-Tune to guide the agent creation flow.
Build natively on watsonx Orchestrate using the ADK, Langflow, or Flow Builder. Or bring existing LangGraph agents directly into the platform. IBM supports the frameworks developers already use.
Whether you built your agent on watsonx Orchestrate or somewhere else entirely, you get the same enterprise-grade security, governance, and observability. No rewrite required to go to production.
Long-term memory, platform credentials, policy enforcement, and evaluations are available to every agent on the platform—native or imported. The control plane treats every agent equally regardless of its origin.
Native agents, imported LangGraph agents and external A2A agents all appear in a single unified dashboard. One control plane to monitor, manage, and govern your entire agent portfolio at an enterprise scale.
Connect your tools, models and data into governed agentic workflows