LangGraph
Open-source framework from LangChain for building AI agents as a graph of individual steps, with saved state and human approvals.
What it does
LangGraph is a library for Python and JavaScript that lets you define an AI agent’s flow as a graph: nodes are work steps, edges decide what happens next. State is saved, so an agent resumes at the same point after an error or a pause. You can add human approvals, stream intermediate results, use memory and have several agents work together. The framework is not tied to any language model and also works without LangChain.
Who it suits
Development teams who want to build their own agents with a tightly controlled flow, for example for support, research or internal processes. Without programming skills you won’t get far with it.
Data protection for businesses
If you run LangGraph yourself, the data stays with you; the terms of the model provider you connect then apply. For hosting and monitoring, LangChain offers the LangSmith platform, optionally with data storage in the EU. According to its terms of service, LangChain does not use customer data for training. A data processing agreement is part of the terms. Self-hosted and hybrid setups are available in the Enterprise plan. LangChain has no contracting entity in the EU.
Limits
LangGraph is a tool for code, not a finished application. Deploying via LangSmith requires the Plus plan and is billed additionally by compute time. The chosen region cannot be changed later.