LangGraph is an open-source library from LangChain, Inc. for building stateful, multi-actor agentic applications. Where LangChain provides primitives for LLM calls and tool use, LangGraph adds a graph execution engine that supports cycles, branching, persistence, and human-in-the-loop interrupts — the infrastructure needed for production-grade agentic workflows.
Key capabilities
Graph-based execution — Workflows are defined as directed graphs where nodes are functions (or LLM calls) and edges are transitions. Cycles are supported natively, enabling the retry and reflection patterns common in agentic systems.
Stateful persistence — LangGraph checkpoints graph state to a configurable store (in-memory, SQLite, Postgres). This enables long-running workflows to pause, resume, and recover from failures.
Human-in-the-loop — Graphs can declare interrupt points where execution pauses and waits for human approval or input before continuing. This is a first-class primitive, not a workaround.
Multi-agent patterns — LangGraph supports Supervisor patterns (one agent delegates to specialists), hierarchical agents, and agent handoffs via send(). The same graph engine runs both the orchestrator and each sub-agent.
LangGraph Cloud / Studio — A hosted runtime and visual debugger for LangGraph graphs. Studio lets developers replay runs, inspect state at each node, and edit checkpoints interactively.
Autonomy level
Level 3 (supervised agent by default): the graph designer controls where humans are required. Graphs can be configured for fully autonomous operation (level 4-5) or tight human oversight (level 1-2) depending on the interrupt points declared.
Strengths
- Fine-grained control over agent behavior vs. higher-level abstractions
- Strong persistence story for long-running workflows
- Model-agnostic; works with any LangChain-compatible LLM
- Open-source core with active community
Limitations
- Lower-level than agent frameworks like AutoGen; requires more code to express common patterns
- Python-first (TypeScript port exists but lags behind)
- LangGraph Cloud is paid; self-hosting requires infrastructure work