AI models and agent platforms · Agent build and run platforms
LangGraph and LangSmith
Open-source framework for controllable, stateful agents, with LangSmith for tracing, evaluation and deployment.
- Vendor
- LangChain
- Pricing model
- Framework free (MIT). LangSmith per seat plus traces and deployment usage.
- Runs on
- Self-hosted anywhere, or LangSmith cloud, hybrid or self-hosted
- Agentic OS layer
- Agents
- Industries
- Any industry
- Last reviewed
- November 2026
Our verdict
The most popular way to build agents your engineers fully control: explicit steps, human approvals, retries and memory. The framework is free; the cost is engineering time and, if you use it, LangSmith for tracing and deployment.
What it is
LangGraph is an open-source framework for building agents as graphs: explicit steps, branches and loops, with state saved at every step. That makes agents predictable, resumable after failure, and easy to pause for human approval, which is what production business agents need.
LangSmith is LangChain's paid platform around it: tracing every step an agent takes, evaluating quality, managing prompts and deploying agents. It works with LangGraph and other frameworks. LangGraph itself runs anywhere, including on AWS AgentCore, Google Cloud or your own servers.
Best for
Engineering teams building multi-step agents for finance, operations or service where every step must be visible, testable and approvable, and organisations that want to avoid lock-in to one cloud or model.
Not the right fit
Business teams without engineers, and simple single-step assistants where a suite tool is faster.
Key capabilities
Graph-based control
Explicit nodes and edges for steps, branches, loops and multi-agent hand-offs.
Durable state
Checkpoints at every step, so runs resume after failures or restarts.
Human in the loop
Pause an agent for approval or edits at any step, then continue.
Streaming and memory
Stream progress to users and keep short- and long-term memory.
LangSmith tracing and evaluation
See every step, cost and latency; test agents against datasets before release.
Deployment options
Managed cloud, hybrid, or fully self-hosted for enterprise data rules.
Strengths
- Open source, model-agnostic and cloud-neutral
- Fine control over every step and approval
- Large community, examples and integrations
- LangSmith gives clear traces and evaluations
Watch-outs
- Needs strong engineers; there is no business-user builder
- Hosted deployment bills per node execution and uptime, which adds up for looping agents
- Trace volume can raise LangSmith costs quickly
- Frequent releases mean upgrades need testing
Pricing
LangGraph is free under the MIT licence with no usage limits. LangSmith has a free Developer plan (one seat, about 5,000 traces a month), a Plus plan at $39 per seat per month with about 10,000 base traces and one small free deployment, and custom Enterprise pricing with hybrid and self-hosted options.
Beyond the base allowance, traces bill per thousand and hosted deployments bill on node executions and standby time. An agent that loops through a tool five times pays for five node executions per request.
Checked November 2026. Vendors change plans often, so confirm current pricing with LangChain before you budget.
Security and governance
Self-host the framework and LangSmith inside your own network for full data control, or use the managed cloud with SSO and role-based access on Enterprise. Security depends on how you deploy and which model provider you call.
How we would roll it out
- Map the process as a graphDraw each step, decision and approval before writing code; that drawing becomes the graph.
- Add approvals where money movesPut human-in-the-loop interrupts before payments, postings, contract changes and customer messages.
- Trace and evaluate from day oneSend traces to LangSmith or another tool and build an evaluation set from real cases.
- Choose where it runsDeploy on your cloud, AgentCore or LangSmith based on data rules, not convenience.
Compare with
Other tools in AI models and agent platforms, closest match first.
Microsoft Copilot Studio
Agent build and run platforms
Low-code agent builder for Microsoft 365, Teams, Dynamics and Power Platform, billed on Copilot Credits.
Best for Microsoft 365 organisations building internal and customer-facing agents.
Amazon Bedrock AgentCore
Agent build and run platforms
Modular AWS services to run, connect, secure and observe production agents, with any framework or model.
Best for Engineering teams running production agents inside AWS.
ChatGPT Enterprise
Enterprise AI assistants
The most widely adopted general AI assistant, with enterprise admin, residency and compliance controls.
Best for Organisations that want one assistant everyone already knows how to use.
Claude Enterprise
Enterprise AI assistants
Assistant, coding agent and desktop task agent in one seat, with published seat pricing and usage billed separately.
Best for Teams that work with long documents, write a lot, or want coding and desktop agents in the same seat.
Gemini Enterprise
Enterprise AI assistants
Google's agent platform for every employee: enterprise search, prebuilt agents and a no-code agent builder.
Best for Google Workspace organisations that want search and agents across company data.
Choosing between LangGraph and LangSmith and the alternatives?
Book a call. We look at your systems, data and governance together and tell you which platform fits, and what to build on it first.