LangGraph
Agents written as an explicit state graph.
- made by
- LangChain
- source
- Official LangGraph site
Magic Ship is one shop in Vancouver, BC, working remotely with clients worldwide. We are not a partner, reseller, or certified vendor of LangChain - we just build with this.
What LangGraph is
LangGraph models an agent as nodes and edges over a typed state object rather than as one prompt loop. It has checkpointing, so a run can pause for human approval and resume later, plus streaming, retries, and branching. Python and TypeScript versions exist and it works with any model provider.
How we use it
We use it when the control flow is the requirement: a step that must be approved by a person, a branch that only runs for one document type, a retry policy that differs per node. Checkpoints go in Postgres so a paused run survives a deploy. Because the graph is explicit, the diagram in the design doc and the code stay honest with each other.
Where it is the wrong choice
For a straightforward call-tools-until-done loop the graph is ceremony, and the LangChain abstractions above it add indirection you have to read through when something breaks. The provider SDK is less code for that case.
Service lines it turns up in
Related tools
More in Agent frameworks
Other tools in the same service lines
Building something on LangGraph?
Send the problem rather than a job spec. You get an answer on scope, on fit, and on whetherLangGraph is even the right call for it.
Start a projectLangGraph and LangChain are trademarks of their respective owners, used here to say what we work with.