fruitpunch

Review · AI & data

LangSmith (LangGraph) logoLangSmith (LangGraph) review

LangChain's platform for tracing, evaluating and deploying agents built with LangGraph or other frameworks.

By the FruitPunch research desk · Updated November 20, 2026

The short answer

LangSmith is LangChain's paid platform for tracing, evaluating and running AI agents, and LangGraph is the free, MIT-licensed framework many teams use to build those agents. It is a strong pick for engineering teams shipping custom agents: the Plus plan listed at $39 per seat per month with 10k base traces and one small serverless Agent Server included when checked in November 2026. Teams without developers should look at a no-code agent builder instead.

Quality (40%)
8.7
Ease of use (25%)
6.4
Value (35%)
7.8

What it looks like

LangSmith (LangGraph) screenshot: LangChain's LangSmith Agent Server page, describing the managed runtime for deployed agents
LangChain's LangSmith Agent Server page, describing the managed runtime for deployed agents. Source: official page (langchain.com/langsmith/deployment), November 20, 2026
LangSmith (LangGraph) screenshot: LangSmith pricing page with the Developer, Plus and Enterprise plans
LangSmith pricing page with the Developer, Plus and Enterprise plans. Source: official page (langchain.com/pricing), November 20, 2026
LangSmith (LangGraph) screenshot: LangSmith Observability page covering tracing, monitoring and Insights
LangSmith Observability page covering tracing, monitoring and Insights. Source: official page (langchain.com/langsmith/observability), November 20, 2026

Our verdict

LangSmith suits developers and AI engineering teams that write their own agent code, in LangGraph or another framework, and need tracing, evals and a managed runtime in one account. It is a poor fit for business users who want to build agents without code, and for small teams that need self-hosting, custom SSO or an SLA, since those sit in the custom-priced Enterprise plan. The deciding fact is how the bill is built: a $39 seat fee plus usage in LangChain Standard Units at $1.00 each, with extra base traces at 0.005 LSU each (about $5 per 1,000), so cost tracks trace volume and deployment size rather than seat count alone.

Overview

LangSmith is the commercial platform from LangChain, the company behind the open-source LangChain and LangGraph libraries. It bundles several services under one account: Observability (tracing, dashboards, alerts), Evaluation (datasets, offline and online evals, annotation queues), Agent Server (the managed runtime for deployed agents), plus newer add-ons called Fleet, Engine, Sandboxes and LLM Gateway. LangGraph itself is described on LangChain's site as an MIT-licensed open-source library that is free to use; the paid part starts when you trace, evaluate or host agents on LangSmith.

The product line was renamed and repriced recently. The former LangGraph Platform became LangSmith Deployment, and LangChain's Agent Server page says LangSmith Deployment was renamed to LangSmith Agent Server as of November 2026. LangChain's billing docs state that usage-based pricing replaced per-run and uptime pricing on October 1, 2026: deployments are now billed on requested vCPU and memory, converted to LangChain Standard Units (LSUs). Serverless deployments scale to zero and cost 0 LSU while scaled down; Dedicated deployments are always on.

On price, LangChain's pricing page lists three plans. Developer is free for one seat with up to 5k base traces a month. Plus costs $39 per seat per month, includes 10k base traces, access to Agent Server and Engine, unlimited seats and one small serverless Agent Server at no charge. Enterprise is custom priced and adds self-hosted and hybrid deployment, custom SSO, ABAC and RBAC, and a support SLA. Base traces are kept 14 days; extended traces are kept 180 days for an extra fee.

Compared with the other agent platforms on this site, LangSmith is the most code-first. Zapier Agents, Make, Lindy and MindStudio target people who build in a visual editor. n8n and Dify offer self-hosting on their open-source editions, while LangSmith's self-hosted and hybrid options require Enterprise. CrewAI is the closest code-first rival, with a free plan capped at 50 executions a month and no self-serve paid tier. LangSmith's tracing is framework-agnostic: its observability page lists SDKs for Python, TypeScript, Go and Java and OpenTelemetry support.

The limits are real. There is no no-code builder for the core product, though Fleet adds agents created in everyday language. Plus is billed monthly with no annual discount shown, and the seat fee covers only the allowance: traces beyond 10k, extra Agent Servers, Engine runs (about 7 to 45 LSUs per run, scheduled every 6 hours) and Sandboxes beyond the 8 LSU allowance all add usage charges. A Dedicated Small deployment has a minimum cost of 0.5335 LSU per hour, which is about $390 over a 730-hour month before any traffic.

How it scored

LangSmith, where the former LangGraph Platform now lives as LangSmith Deployment and Agent Server, is the developer choice on this list. Agent Server runs agents with durable execution, streaming, human-in-the-loop approvals, a memory store, crons, task queues and webhooks, and it supports A2A, MCP and Agent Protocol. The Studio IDE helps debug agent runs visually. Pricing starts with a free Developer plan for one seat and 5k base traces a month. Plus costs $39 per seat per month, includes 10k base traces and one small serverless Agent Server, and bills extra usage in LangChain Standard Units at $1.00 each. Self-hosted and hybrid deployment, custom SSO and an SLA are Enterprise features. It ranks lower here only because it expects you to write code; for a business without engineers, Zapier Agents or Make is a better fit.

Who it is for

developers shipping custom agents to production

Good fit if

  • Developers building agents in LangGraph or LangChain who want tracing, evals and one-click deployment from the same vendor
  • Teams using other frameworks such as the OpenAI SDK, Anthropic SDK, Vercel AI SDK, LlamaIndex, CrewAI or Google ADK that need framework-agnostic tracing and a managed runtime
  • AI engineering teams that run eval suites in CI with pytest, Vitest or GitHub workflows and want to fail builds when scores drop
  • Solo developers who can stay within one seat and 5k base traces a month on the free Developer plan
  • Enterprises that need agents and traces kept inside their own VPC through hybrid or self-hosted deployment

Skip it if

  • Business users who want to build agents without code: Zapier Agents or Make offer visual builders
  • Small teams that want to self-host the whole stack without an enterprise contract: n8n's free Community edition or Dify's open-source edition run on your own server
  • Teams focused on retrieval-based chat apps and knowledge bases with a visual editor: Dify is the closer match
  • Organizations standardized on Microsoft 365 that want agents inside Teams and Copilot: Microsoft Copilot Studio fits that stack

Specifications

From LangSmith (LangGraph)'s own pages, each linked

Free plan
Developer: 1 seat, up to 5k base traces per month, community support
source
Cheapest paid plan
Plus: $39 per seat per month, then pay as you go
source
Usage unit
LangChain Standard Unit (LSU), 1 LSU = $1
source
Extra traces
0.005 LSU per base trace; 0.0025 LSU to upgrade a trace to extended retention
source
Trace retention
Base traces 14 days; extended traces 180 days
source
Included deployment
1 small serverless Agent Server on Plus at no charge
source
Serverless Agent Server sizes
Small 0.8 vCPU / 2 GiB at 0.072 LSU per hour while running; Medium 0.306 LSU; Large 0.612 LSU
source
Dedicated Agent Server minimum
Small 0.5335 LSU per hour, Medium 1.067 LSU, Large 2.134 LSU (runtime plus database)
source
Agent protocols
A2A, MCP and Agent Protocol endpoints; agents can be exposed as an MCP server
source
Tracing SDKs
Python, TypeScript, Go and Java, plus OpenTelemetry
source
Frameworks for deployment
LangGraph and LangChain natively; Google ADK, Claude Agent SDK, Strands, CrewAI and AutoGen through wrappers
source
Hosting options
Cloud on Developer and Plus (US or EU); Cloud, Hybrid or Self-Hosted on Enterprise
source
State payloads
Persistent checkpoints with up to 25 MB payloads
source
Training on customer data
No, per LangChain's pricing FAQ
source
LangGraph license
MIT, free to use
source

Price

Free Developer plan for 1 seat; Plus $39 per seat per month then pay as you go.

Plus For teams building and deploying agents. $39 / seat per month then pay as you go
Quoted from the vendor pricing page on November 20, 2026.

Plans and prices

Quoted from the seller, with the date we checked

LangSmith (LangGraph) plans and prices with sources and check dates
PlanPriceWhat is includedChecked
Developer$0 per seat per month, then pay as you gomonthly, self-serve; usage billed in arrears1 seat; Up to 5k base traces per month; Fleet with 7 LSU included per month; Sandboxes with 8 LSU included per month; Community support; Google and GitHub sign-inChecked Nov 20, 2026
Plus$39 per seat per month, then pay as you gomonthly, self-serve; seats billed on the 1st or pro-rated, usage billed in arrearsUnlimited seats at $39 each; Up to 10k base traces per month; 1 small serverless Agent Server at no charge; more billed by resource usage; Access to Agent Server, Engine and Tuned Evaluators (US public beta); Fleet with 37 LSU included per month; Sandboxes with 8 LSU; Email support; organization roles (User and Admin)Checked Nov 20, 2026
EnterpriseCustom pricing, then pay as you goannual invoice, custom termsCloud, Hybrid or Self-Hosted deployment; Custom SSO, ABAC and RBAC; Support SLA, Slack support, team trainings and architectural guidance; Custom seats and workspaces; Infosec reviewChecked Nov 20, 2026
Startup planDiscounted rates; up to $10,000 in creditsby application, for VC-backed early-stage companiesDiscounted rates; Larger free trace allotments, per LangChain's pricing FAQChecked Nov 20, 2026

Key features

  1. Tracing and monitoring

    LangSmith records each agent run as a trace with every LLM call and tool step. Dashboards track token usage, P50 and P99 latency, error rates, cost and feedback scores, with webhook and PagerDuty alerts. LangChain says the SDK sends traces asynchronously, so an outage on its side does not stop the agent.

  2. Offline and online evaluation

    Datasets support offline evals before release, and online evals score production traffic. Evaluator types include human annotation queues, heuristic checks, LLM-as-judge and pairwise comparison, plus custom evaluators in Python or TypeScript. Evals can run in CI through pytest, Vitest or GitHub workflows.

  3. Agent Server runtime

    The managed runtime provides durable execution, streaming, human-in-the-loop approvals, a memory store with semantic search, cron jobs, double-texting handling, a task queue and webhooks, with 30+ API endpoints. It is framework-agnostic: agents built outside LangGraph can be wrapped and deployed.

  4. Serverless or Dedicated hosting

    Serverless deployments scale to zero and are billed only while running, with cold starts and one replica. Dedicated deployments run at least 2 API server replicas and 1 queue worker with their own database, and autoscale. LangChain recommends Dedicated for customer-facing agents.

  5. Studio IDE

    Studio is a visual debugger for agent graphs. LangChain's docs say it runs in the browser and connects directly to a local Agent Server, so the data sent to that server is not sent to LangSmith.

  6. Protocol support

    Agent Server exposes A2A, MCP and Agent Protocol endpoints, and a deployed agent can be exposed as an MCP server. An agent registry adds versioning and instant rollbacks.

  7. Engine and Insights

    Engine runs every 6 hours to cluster traces into issues, diagnose failures and propose prompt or code fixes with eval coverage, at about 7 to 45 LSUs per run per LangChain's FAQ. Insights clusters traces by topic and produces error analysis summaries.

  8. Spend controls

    Workspace spend limits and trace retention settings let teams cap usage. The billing docs advise setting serverless scale-down to 1 hour instead of the default 1 week and running several graphs on one deployment to share replicas.

Strengths and weaknesses

What we like

  • Free Developer plan includes 5k base traces a month, enough to instrument a prototype before paying anything
  • Plus at $39 per seat includes 10k base traces and one small serverless Agent Server, so a small team can deploy an agent without a separate hosting bill
  • Extra base traces cost 0.005 LSU each, about $5 per 1,000, which is easy to forecast against volume
  • Framework-agnostic: tracing SDKs for Python, TypeScript, Go and Java plus OpenTelemetry, and deployment wrappers for Google ADK, Claude Agent SDK, CrewAI and AutoGen
  • Serverless deployments cost 0 LSU while scaled down, and a running Small serverless deployment costs 0.072 LSU per hour
  • LangGraph, the framework underneath, is MIT-licensed and free, so the agent code itself is not locked to a paid plan
  • LangChain states it does not train models on traces, prompts or outputs sent to LangSmith

What could be better

  • No visual agent builder for the core platform; building agents means writing code
  • Self-hosted and hybrid deployment, custom SSO, RBAC beyond User and Admin roles, and an SLA all require the custom-priced Enterprise plan
  • A Dedicated Small deployment has a minimum cost of 0.5335 LSU per hour, about $390 over a 730-hour month, even when idle
  • Base traces are kept only 14 days; 180-day retention costs an extra 0.0025 LSU per trace
  • The Developer plan is limited to 1 seat and personal organizations cannot invite users, so any collaboration means paying for Plus
  • Several meters (traces, deployments, Engine, Sandboxes, Fleet, Gateway) make the monthly bill harder to predict than a flat plan

Key facts

  • Free plan: Developer, 1 seat, 5k base traces per month
  • Cheapest paid plan: Plus, $39 per seat per month plus usage
  • Included deployment: 1 small serverless Agent Server on Plus
  • Usage unit: LangChain Standard Unit, $1.00 each
  • Protocols: A2A, MCP, Agent Protocol

Where it ranks

LangSmith (LangGraph) is No. 9 in our Best AI Agent Platforms verdict.

Alternatives to consider

CrewAI

The closest code-first rival, built around multi-agent crews; its free plan includes tracing and MCP export but caps at 50 executions a month, and paid plans are sales-led.

n8n

A visual workflow tool with AI agent nodes and a free self-hosted Community edition, for teams that want self-hosting without an enterprise contract.

Dify

An open-source LLM app builder with a visual editor and knowledge bases, better for retrieval-based chat apps built by mixed technical teams.

Zapier Agents

A no-code option for business users who want agents connected to everyday apps without writing or deploying code.

Microsoft Copilot Studio

Fits organizations already on Microsoft 365 that want agents published into Teams and Copilot with Microsoft's admin controls.

Questions readers ask

Is LangSmith free?
Yes, the Developer plan is $0 for 1 seat with up to 5k base traces a month. After a credit card is added, extra traces are billed at 0.005 LSU each, and Agent Server hosting requires the Plus plan or above.
How much does LangSmith cost?
Plus costs $39 per seat per month, then pay as you go, per LangChain's pricing page in November 2026. It includes 10k base traces and one small serverless Agent Server; extra usage is billed in LangChain Standard Units at $1.00 each. Enterprise is custom priced.
Is LangGraph free?
Yes, LangGraph is an MIT-licensed open-source library and free to use. Costs only apply if you trace, evaluate or host agents on LangSmith, or pay for your own infrastructure.
What happened to LangGraph Platform?
It is now LangSmith Agent Server. LangGraph Platform was first folded into LangSmith Deployment, and LangChain says LangSmith Deployment was renamed LangSmith Agent Server as of November 2026.
Do I have to use LangChain or LangGraph to use LangSmith?
No, LangSmith is framework-agnostic. LangChain says it traces apps built with the OpenAI SDK, Anthropic SDK, Vercel AI SDK, LlamaIndex or custom code, and Agent Server can deploy agents from Google ADK, Claude Agent SDK, CrewAI and AutoGen through wrappers.
Can LangSmith be self-hosted?
Yes, but only on the Enterprise plan, which offers Hybrid (LangChain control plane, your data plane) and fully Self-Hosted options. Developer and Plus run on LangChain's cloud in the US or EU.
How long does LangSmith keep traces?
Base traces are kept for 14 days and extended traces for 180 days. Upgrading a base trace to extended retention costs an extra 0.0025 LSU per trace.
LangSmith vs CrewAI: which should a developer pick?
LangSmith has a self-serve paid plan at $39 per seat with a managed runtime, while CrewAI's paid tier is Enterprise only. CrewAI's free plan covers 50 executions a month and suits multi-agent prototypes; LangSmith suits teams that need tracing and deployment at production volume.

Sources we checked

  1. LangSmith Plans and Pricing, checked November 20, 2026
  2. LangSmith Agent Server, checked November 20, 2026
  3. LangSmith Observability, checked November 20, 2026
  4. LangSmith Evaluation, checked November 20, 2026
  5. LangGraph, checked November 20, 2026
  6. LangSmith docs: Manage billing in your account, checked November 20, 2026
  7. LangSmith docs: Deploy agents, checked November 20, 2026
  8. LangSmith docs: LangSmith Agent Server, checked November 20, 2026
  9. LangSmith docs: Data storage and privacy, checked November 20, 2026
  10. LangSmith docs: Organization overview, checked November 20, 2026

Deep review last updated November 20, 2026. We recheck prices and specs against these pages; send corrections to the address below.

Alternatives we scored

Visit langchain.com · Questions or corrections: editorial@fruitpunch.ai