LangGraph

What is LangGraph?

LangGraph is an open-source framework and runtime for building stateful, long-running AI agents with explicit control over orchestration. Developers define nodes, state, transitions, and stopping conditions as a graph, which supports deterministic application logic alongside model-driven decisions. It can represent single-agent loops, multiple collaborating agents, hierarchical teams, and workflows that pause for human decisions.

Persistence checkpoints save graph state as execution proceeds. This enables conversational memory, human-in-the-loop review, resuming after interruption, time-travel debugging, and fault tolerance without restarting an entire task. Streaming can expose intermediate state and outputs while an agent is working.

How to Use LangGraph

1. Define the state shared across the workflow.
2. Implement nodes for model calls, tools, retrieval, validation, and human review.
3. Connect nodes with fixed or conditional transitions.
4. Add a checkpointer when the workflow must pause, resume, remember, or recover.
5. Stream intermediate events and inspect state during testing.
6. Add authorization, tool limits, idempotency, and evaluation before production use.

Core Features

• Graph orchestration: Models agent behavior as explicit nodes, state, and transitions.
• Durable execution: Saves progress so long-running work can resume after interruption.
• Persistence and memory: Maintains thread state and application-defined long-term context.
• Human-in-the-loop: Pauses before sensitive actions and resumes with reviewer input.
• Multi-agent patterns: Supports supervisors, teams, handoffs, and hierarchical flows.
• Streaming: Emits intermediate steps, messages, and custom events.
• Fault tolerance: Uses checkpoints and pending writes to recover safely.
• Low-level control: Allows custom routing instead of forcing a fixed agent loop.

Use Cases

• Build an agent that pauses for approval before changing an external system.
• Coordinate specialist agents across research, analysis, and writing.
• Run long tasks that must survive process or network interruptions.
• Maintain state across a conversation or business case.
• Debug complex workflows by inspecting and replaying checkpoints.

Pricing

LangGraph is open-source software and can be used without a framework license fee. Managed tracing, evaluation, and deployment are available separately through LangSmith. Its Developer plan is $0 per seat per month, Plus is $39 per seat per month plus usage, and Enterprise pricing is custom. Model and infrastructure costs are separate.

Frequently Asked Questions

Is LangGraph the same as LangChain?

No. LangGraph provides lower-level stateful orchestration, while LangChain offers higher-level agent abstractions and integrations that can run on LangGraph.

Why are checkpoints important?

They preserve state for memory, approvals, recovery, replay, and long-running execution.

Does durable execution make tool calls automatically safe?

No. External actions still need permissions, validation, idempotency, approval rules, and failure handling.