LangGraph

What it does

LangGraph is a low-level orchestration framework and runtime for building agents. Developers define a graph of steps, mixing hand-coded deterministic logic with LLM-driven steps. Each run carries shared state, and the library returns agent outputs, streamed intermediate steps, and persisted state and memory that can be resumed or inspected.

Developers install it with pip, define nodes and edges around a shared state, compile the graph and invoke it. Typical uses include single-agent, multi-agent and hierarchical control flows, human approval checkpoints, and agents that resume after failures. It does not abstract prompts or architecture, so you decide how the agent behaves. It can be used without LangChain, though LangChain components are used throughout the documentation.

It suits developers who need fine-grained control over production agents and are comfortable writing code; teams wanting a higher-level starting point are pointed to LangChain's prebuilt agents. The library is MIT-licensed and free, but you supply your own model and tools. Tracing, evaluation and managed deployment come from LangSmith, a separate product with its own paid plans.

What it can do

• Mix deterministic and LLM-driven steps in one graph
• Persistence so agents resume after failures
• Human-in-the-loop inspection and modification of agent state
• Short-term and long-term memory across sessions
• Token-by-token streaming of agent reasoning and actions
• Single, multi-agent and hierarchical control flows

Works with

• LangChain
• LangSmith

Example workflows

Approval-gated support agent

Input: a customer request and a set of tools. Steps: define a graph with an LLM step and a deterministic action step, add a human-in-the-loop checkpoint before sensitive actions, and stream progress. Output: an approved or corrected action with conversation memory kept for later sessions.

Long-running task that resumes

Input: a multi-step job such as research or data processing. Steps: model each stage as a node with persisted state, run it, and resume from the saved state after any interruption. Output: a completed result plus an inspectable record of state transitions.

Before you choose

• It is very low-level and focused on orchestration, so you assemble models, tools and prompts yourself.
• The library is free, but LangSmith tracing, deployment and other services are metered separately and are not required to run it.

Listed under

• Coding
• Agent framework
• Agent memory
• Agent orchestration
• Developers
• Developer tools
• Human-in-the-loop

Get started

pip install -U langgraph; source on GitHub

1. Install with pip install -U langgraph (or uv add langgraph).
2. Build a StateGraph with nodes and edges, compile it and invoke it, as in the hello-world example.
3. Optionally set LANGSMITH_TRACING=true and your API key to trace and debug runs.

The MIT-licensed library is free to use. Model-provider, infrastructure and optional LangSmith services have separate charges.

Source: langchain.com/langgraph, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• LangGraph: Agent Orchestration Framework for Reliable AI Agents ↗www.langchain.com
• LangGraph overview - Docs by LangChain ↗docs.langchain.com
• LangSmith Plans and Pricing ↗www.langchain.com
• Home - Docs by LangChain ↗docs.langchain.com

Company website

langchain.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.