LangGraph

LangGraph is LangChain's low-level orchestration framework and runtime for long-running, stateful agents. Official docs say it is trusted by Klarna, Uber, J.P. Morgan, and others, and that you mix deterministic, hand-coded steps with LLM-driven steps in one graph. It is focused on orchestration: durable execution, streaming, human-in-the-loop, and persistence. You do not have to use LangChain to use LangGraph. If you want a prebuilt tool loop, docs now point you at LangChain agents first. The repo langchain-ai/langgraph had about 39,835 stars on 2026-08-17.

Key Features

• Graph runtime: StateGraph, nodes, edges, START/END, then compile() and invoke().
• Durable + streaming: Agents persist through failures and stream state.
• Human-in-the-loop: Inspect and change state mid-run.
• Stack around it: Deep Agents (harness), LangChain (models/tools/loops), LangSmith (trace/eval/deploy), LangSmith Engine (fix proposals from traces), LangSmith Fleet (no-code builder).
• Install: pip install -U langgraph or uv add langgraph.

Use Cases

• Bespoke agents that must be auditable on some steps and flexible on others.
• Long jobs that resume after a crash.
• Teams already on LangSmith who need the runtime under Deep Agents.

Pricing

LangGraph open source is free (check the repo license on GitHub). LangSmith tracing, deployment, Engine, and Fleet are separate LangChain Cloud products with their own plans. Confirm LangSmith pricing before you treat Studio deploy as free.

Getting Started

1. Read the LangGraph overview.
2. pip install -U langgraph and run the hello-world StateGraph in that page.
3. Add LangSmith tracing before you debug a real graph.
4. Use LangChain's prebuilt agents if you only need a default tool loop.

Frequently Asked Questions

Is LangGraph the whole LangChain product?

No. Docs split harness (Deep Agents), framework (LangChain), runtime (LangGraph), and platform (LangSmith).

Python only?

The page we used is the Python OSS overview. Confirm current JS/TS docs on the same docs site before you promise a second language.

39k stars still right?

About 39,835 on 2026-08-17. Confirm the API.

Alternatives

• LangChain: Higher-level models, tools, and agent loops.
• OpenAI Agents SDK: Official OpenAI primitives without a graph runtime.
• Dify: Visual workflows if you do not want to write graphs.

Tips

1. Do not start in LangGraph if you only need a chatbot with tools. Use LangChain agents, then drop to a graph.
2. Turn on checkpoints before the first overnight run.
3. Confirm the products comparison page. Names (Engine, Fleet) are new.