Technology RadarTechnology Radar

LangGraph

multi-agentframework
This item was not updated in last three versions of the Radar. Should it have appeared in one of the more recent editions, there is a good chance it remains pertinent. However, if the item dates back further, its relevance may have diminished and our current evaluation could vary. Regrettably, our capacity to consistently revisit items from past Radar editions is limited.
Trial

LangGraph is a graph-based framework for building stateful, multi-step AI agent workflows — the recommended successor to LangChain for anything involving agents or complex pipelines.

Buy vs Build

LangGraph is a build tool — you write Python code that defines your agent workflow as a graph. If you want a managed hosted version, LangSmith (by the same team) provides observability and LangGraph Platform offers hosting. But the framework itself is open-source.

Why It's in Trial

LangGraph is in production at LinkedIn, Uber, and 400+ other companies. The LangChain team officially recommends it over LangChain for agent orchestration: "Use LangGraph for agents, not LangChain." Most notably, LangChain's own Open SWE — their open-source autonomous coding agent framework — is built entirely on LangGraph, validating it as the foundation for production multi-agent systems.

What It Actually Is

Think of your agent workflow as a flowchart — LangGraph lets you define it as code:

  • Nodes: A function or tool call (e.g., "call the LLM", "search the web", "write to database")
  • Edges: The connections between nodes, which can be conditional ("if the LLM says it needs more information, go to the search node")
  • State: A shared data structure that persists across the entire workflow — so each node can see what previous nodes decided

This graph-based approach makes it much easier to:

  • Debug long-running agent tasks (you can see exactly where it went wrong)
  • Add human-in-the-loop checkpoints (pause and wait for approval before continuing)
  • Build branching workflows that react differently to different situations

When to Choose LangGraph

  • Building custom multi-agent systems where you need precise control over flow
  • Complex workflows with conditional branching, loops, or error recovery
  • When you need fine-grained observability into what the agent is doing

Getting Started

pip install langgraph

Then model your workflow as a graph where nodes are Python functions. The state flows through the graph as the agent executes.

Key Characteristics

Property Value
Language Python (primarily)
License MIT
Hosted option LangGraph Platform (managed)
Provider LangChain Inc.
Website langchain-ai.github.io/langgraph
GitHub langchain-ai/langgraph

Further Reading