Buying snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Decision point | What LangChain states |
|---|---|
License | MIT-licensed open-source library, free to use |
Installation | Python package installed with pip install -U langgraph; LangGraph.js is the equivalent JS/TS library |
Managed hosting | Optional, through LangSmith Deployment, a separate paid LangSmith service |
Self-hosted LangSmith | Self-hosted and hybrid deployment options are listed on the Enterprise plan; Hybrid means a SaaS control plane with a self-hosted data plane |
Model training on your data | LangSmith does not use your traces, prompts or outputs to train models |
Our take
Pick LangGraph if your engineers want to define agent control flow in their own Python or JS code and run it on their own infrastructure, since the library itself costs nothing. Budget for a LangSmith plan only if you also want LangChain to host, trace or evaluate those agents.
Evaluate LangGraph
Popular comparisons
See all 7 LangGraph comparisons
Where it stands out
Durable execution. Agents persist through failures and resume from exactly where they left off, which matters for long-running work that should not restart from zero.
Human approval points. You can inspect and modify agent state at any point during execution and add checks that steer and approve agent actions before they happen.
One framework for many shapes. Low-level primitives cover single-agent, multi-agent and hierarchical control flows, so a team does not have to switch tools as designs grow.
Memory and streaming built in. It keeps short-term working memory and long-term memory across sessions, and streams output token by token so users can see agent reasoning and actions in real time.
Product screenshot
Trade-offs to weigh
Low-level by design. LangChain itself points teams that want to build agents quickly to Deep Agents, a higher-level package built on LangGraph for planning, subagents and file systems. Expect to write more of the agent logic yourself with plain LangGraph.
Dedicated hosting bills while live. On LangSmith Deployment, uptime runs for as long as a deployment's database is live. Serverless deployments scale to zero when idle, but dedicated deployments are always on, and LangChain recommends dedicated for any customer-facing agent.
Short default trace history. Base traces are kept for 14 days. Keeping traces for 180 days means upgrading them to extended traces for an additional fee, which matters if you need a longer record of agent decisions.
Pricing and plan conditions
The LangGraph library is free under the MIT license. The plans below are for LangSmith, LangChain's platform for tracing, evaluation and managed deployment. On Developer and Plus, seats are billed monthly on the 1st, pro-rated when added mid-month, with no credit for removed seats; traces are billed monthly in arrears. Enterprise plans are invoiced annually upfront. Usage beyond plan allowances is metered in LangChain Standard Units (LSUs) at 1 LSU = $1; on Plus, LSUs are charged on top of the monthly seat fee, and on Enterprise they are charged against committed LangChain credits. LangChain also offers VC-backed startups discounted rates and up to $10,000 in credits.
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Plan | Published price | What it adds |
|---|---|---|
LangGraph library | Free | MIT-licensed open-source framework you run yourself |
LangSmith Developer | $0 / seat per month, then pay as you go | 1 seat; up to 5k base traces / mo, then pay-as-you-go; community support; Deployment is N/A |
LangSmith Plus | $39 / seat per month | Unlimited seats; up to 10k base traces / mo, then pay-as-you-go; access to Deployment, Engine and more; 1 free Serverless (Small) deployment included |
LangSmith Enterprise | Custom pricing | Self-hosted and hybrid deployment options; custom SSO, ABAC and RBAC; support SLA; custom seats and workspaces |
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Meter | Rate |
|---|---|
Runtime Compute | 0.0675 LSU / vCPU-hr |
Runtime Memory | 0.0090 LSU / GiB-hr |
Database Compute | 0.177 LSU / vCPU-hr |
Database Memory | 0.025 LSU / GiB-hr |
Public signals
About these signals
Verified factual signals from public sources. They indicate observable activity or interest, not total adoption, product quality, or cost.
208 GitHub commits 90d43.9k GitHub stars0 vulnerabilities across 2 packages
See all signals from 7 sources
npm · @langchain/langgraph@1.5.2 · PyPI · langgraph@1.2.17
Alternatives to LangGraph
The reviewed substitutes for LangGraph among the agent frameworks, and what would make each one the better answer.
Direct alternatives
Reviewed substitutes: products bought for the same job, where a team picks one.
- Agno
- LangGraph and Agno are frameworks for running AI agents reliably in production, each with its own runtime. LangGraph models an agent as a stateful graph with persistence and human-in-the-loop control, while Agno stresses fast agent execution and ships AgentOS, a control plane to chat with, trace and monitor agents in your own cloud.Applies to: Choosing between these two for the ai agent frameworks decision.
- AutoGPT
- An autonomous agent that works through a task step by step can be built with LangGraph or with AutoGPT. LangGraph gives developers low-level control over agent state, cycles and approvals, while AutoGPT is an open-source project geared toward ready-made assistants that automate digital workflows with less custom code.Applies to: Choosing between these two for the ai agent frameworks decision.
- CrewAI
- CrewAI and LangGraph are open-source Python frameworks for orchestrating several AI agents on complex tasks. CrewAI organises agents as a crew of role-playing members who delegate work and adds a visual editor for enterprises, while LangGraph is a lower-level runtime from the LangChain team for stateful agent graphs with cycles, persistence and human-in-the-loop checks.Applies to: Choosing between two products of the same kind for one job.
- Google Agent Development Kit
- Both express agent workflows as graphs and are adopted as the primary framework. ADK adds a managed deployment target; the framework decision is still one or the other.
- OpenAI Agents SDK
- Both are adopted as the primary agent framework for an application. Handoffs and graphs are two ways to express the same orchestration decision.
Other approaches
A different approach to the same problem. Each substitutes only for the workload named beside it.
- AutoGen
- Both build applications where a model calls tools and takes several steps, from different starting points: framework breadth, explicit orchestration, role decomposition or an SDK inside an existing application. Teams pick one to build on, and the choice follows the shape of the workflow rather than a feature count.Applies to: Choosing the framework an agent or multi-step LLM application will be built on.
- Haystack
- Create agentic, context engineered AI systems using Haystack’s modular and customizable building blocks, built for real-world, production-ready applications.Applies to: when the hard problem is durable agent state and complex cyclic control flow rather than retrieval and context-engineering composition.
- MetaGPT
- Both build stateful multi-actor agent applications; LangGraph is graph-based and controllable, MetaGPT is role-based.Applies to: Building stateful multi-actor agent applications.
- Semantic Kernel
- LangGraph is preferred over Semantic Kernel for stateful agent workflows requiring cycles, persistence, human approval, and explicit control flow. We recommend LangGraph over Semantic Kernel when an analytics or data platform team must model branching execution, approvals, and durable state explicitly, accepting the need to design more of the workflow itself.
Related technologies
Normally used together rather than chosen between, so these are not alternatives.
- LangChain
- LangGraph is built by the LangChain team for explicit stateful orchestration and is commonly used with LangChain components, though either runs without the other. The reader's question is whether adding graph orchestration to an existing LangChain application is a replacement or an addition.Applies to: Whether an existing LangChain application needs graph orchestration added, or rewritten.
- LlamaIndex
- Not a substitute, and the extractor proposed it as one. The 2026 buyer guidance is explicit that this pair is used together rather than chosen between: LlamaIndex for ingestion and retrieval, LangGraph for orchestration and agent control flow, which multiple enterprise sources describe as the pattern production teams already follow. LangGraph brings no ingestion or indexing layer, so it cannot stand in for the half of LlamaIndex a reader comes for. Recorded as complementary so it stays off the alternatives grid instead of telling a reader to pick one.Applies to: Retrieval and stateful agent orchestration in one application: LlamaIndex indexes and retrieves, LangGraph runs the multi-step agent over it.