LangGraph
A low-level orchestration framework for long-running, stateful agent workflows. It provides graph control, persistence, and human-in-the-loop mechanisms rather than a finished assistant.
Utility: Frameworks; kind: framework.
Stars: 44k+; Starred in September: 1k+.
Metrics updated: 31 Oct 2026. Profile reviewed: 31 Oct 2026.
Utilities
- Represent agent state and transitions explicitly.
- Add persistence and human interrupts to multi-step workflows.
Who it is for
- Developers who need control over agent execution and recovery.
Requirements
- Install the Python package or use the separate JavaScript implementation.
- Configure models, tools, and persistent state storage for the application.
Limitations
- Durability depends on the application's persistence and side-effect design.
- Framework control does not guarantee correct model decisions.
Cost and privacy notes
- Models, storage, and optional managed deployment/observability can incur separate charges.
- Persisted checkpoints and traces can retain prompts and tool results.
Illustrative example (untested)
Illustrative, not tested: pause a workflow before an external write, inspect its state, and resume after approval.
License
MIT; reviewed the repository LICENSE.