Geekflare
PermissiveMITTested by Geekflare

LangGraph review

Graph-based framework for building long-running, stateful agents in Python and JavaScript

Website GitHub Docs
Data refreshedMethodologyHow we tested
Director of Geekflare

Rity Anand is a director of Geekflare. She has extensive experience managing business technology and a special interest in emerging technologies.

Verdict

LangGraph gives the most control over how an agent moves between steps, and its checkpointing and interrupts are the most mature of the frameworks here. It takes more code than the higher-level options, and the tracing, Studio and production server that make it easy to run all point to LangSmith.

Best for

Teams building long-running agents that must pause, resume and be audited step by step

Not for

Teams that want a quick agent in a few lines or no tie to a vendor's platform for running it

Pros

  • Durable execution with checkpoints in SQLite or Postgres
  • Interrupts that pause a run for a person to approve or edit state
  • Works with any model provider, with or without LangChain
  • Python and JavaScript versions

Cons

  • Low-level, so simple agents take more code than in other frameworks
  • Tracing and Studio depend on LangSmith, a hosted service
  • The production Agent Server is not open source and needs a licence key to self-host
Ask Sparkian

Is LangGraph right for your project?

Describe what you need and AI will check the fit against this review.

LangGraph features and capabilities

Durable execution

Each step is saved by a checkpointer, so a run that fails or waits for days resumes from the last step instead of starting over.

Human review at any step

A graph can stop with an interrupt, show its state to a person and continue with their changes, which suits approvals before an agent acts.

Explicit control flow

You draw the agent as nodes and edges, with loops, branches and subgraphs, so the path it takes is predictable and testable.

Short and long-term memory

Conversation state stays with each thread, and a store keeps facts across sessions, so agents can remember users without extra services.

Local server and Studio

langgraph dev runs your graph as an API on your machine, and LangSmith Studio connects to it to step through runs and edit state.

Any model providerWorks with OpenAI, Anthropic, Google and local models, not one vendor
Yes
Multi-agentSeveral agents can hand off or work together
YesSubgraphs, supervisors and handoffs
MCP toolsCan use tools from MCP servers
YesThrough the langchain-mcp-adapters package
Durable stateSaves progress so a run can resume after a crash or a pause
YesCheckpointers for SQLite and Postgres
Human approvalCan pause for a person to approve or edit before continuing
Yes
TracingShows each step, tool call and token use
PartlyLangSmith cloud, free for 5,000 traces a month
Visual studioA UI to build, run or debug agents
PartlyLangSmith Studio, a hosted page that connects to your server

At a glance

Type
Agent orchestration framework
Stack
Python 3.10+ / TypeScript
Databases
PostgreSQL, SQLite
Licence terms
The LangGraph library is MIT. The Agent Server that serves graphs over HTTP is Elastic License 2.0, and running it yourself in production needs a LangSmith license key.
SSO
Paid only
Roles and permissions
Paid only
Audit logs
Not documented

Free vs paid editions

What you get when you self-host, and what needs a paid plan.

The LangGraph library is free under MIT and includes checkpointing to SQLite or Postgres, interrupts for human review and streaming. LangSmith adds hosted tracing, evaluations, Studio and managed deployment. Its free Developer plan has one seat and 5,000 traces a month, and Plus costs $39 per seat a month with one small deployment. Running the Agent Server on your own servers in production needs an Enterprise licence.

Self-hosting LangGraph

LangGraph needs Python 3.10 or later. The local dev server needs Python 3.11, and a self-hosted Agent Server needs PostgreSQL and Redis.

New cloud server to ready

45 s

Research task, median of 3

24.8 s, 133 MB peak

Version tested

1.2.16 · Oct 19, 2026

  • Installing LangGraph and LangChain's OpenAI-compatible client in a fresh Python container took 20 seconds, and the first answer came back 25 seconds later.
  • The prebuilt agent needed only the model, the two tool functions and a system prompt. All three runs passed, taking 23 to 36 seconds and $0.001 to $0.003 each.
  • Its tracing and Studio are LangSmith, a separate service with a free tier. Without it, the graph returns the full message list, which is enough to see each tool call.
agent.py, after pip install langgraph langchain-openai httpx
# tools.py holds the shared web_search and read_page functions over Geekflare's APIs,
# common.py the model name, OpenRouter key, system prompt and task.
from langchain_core.tools import tool
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent

import tools
from common import BASE_URL, KEY, MODEL, SYSTEM

web_search = tool(tools.web_search)
read_page = tool(tools.read_page)

llm = ChatOpenAI(model=MODEL, base_url=BASE_URL, api_key=KEY, temperature=0, max_tokens=1024)
agent = create_react_agent(llm, [web_search, read_page], prompt=SYSTEM)


def run(task: str) -> str:
    result = agent.invoke({"messages": [("user", task)]}, {"recursion_limit": 20})
    return result["messages"][-1].content
Agent's answerPassedRun 1 of 3 · 24.8 s · $0.0008
Asked: What is the latest stable release of Scrapy, when was it released, and what are two notable changes in it? Use the tools, cite the release notes URL, and answer in under 120 words.

The latest stable release is Scrapy 2.19.0, released 10 September 2026 according to the Scrapy release notes and the GitHub releases page.

Two notable changes:

  1. A new RemoteControl extension lets you inspect and control a running crawl over HTTP.
  2. The default priority queue is now DownloaderAwarePriorityQueue, per the GitHub release summary.

I could only see the release highlights in search snippets, not the full notes, so I haven't verified these against the complete changelog.

LangGraph alternatives

CrewAIPython framework for teams of role-based AI agents and event-driven flowsPydantic AITyped Python agent framework from the Pydantic team, with validated outputs and any modelOpenAI Agents SDKLightweight SDK from OpenAI for agents with tools, handoffs, guardrails and tracingMastraTypeScript framework for agents and workflows, with a local Studio, memory and tracingMicrosoft Agent FrameworkMicrosoft's framework for agents and multi-agent workflows in Python and .NET, the successor to AutoGen and Semantic KernelGoogle ADKGoogle's Agent Development Kit for building, testing and deploying agents in Python, TypeScript, Java and Go