LangSmith
by LangChain
Framework-agnostic platform to observe, evaluate, deploy and continuously improve production AI agents
LangSmith is LangChain's framework-agnostic agent engineering platform, combining tracing and observability, evaluation, managed deployment, sandboxed code execution and an LLM gateway in one system. It is built for engineering teams moving AI agents from prototype to production who need to debug failures, measure quality and govern rollout across an enterprise.
LangSmith is the commercial platform from LangChain, the company behind the LangChain and LangGraph open-source frameworks, and it has expanded well beyond its original role as an LLM tracing tool into what the vendor now calls an agent engineering platform. It is deliberately framework-agnostic: teams can instrument agents built on LangGraph, Deep Agents or entirely custom code, and run them against closed, open or hybrid model stacks. The platform has four functional layers. Observability captures full execution traces for debugging, usage-pattern analysis and monitoring dashboards. Evaluation scores agent behaviour using LLM-as-judge, code-based or multi-turn evaluators applied to real production traces. Infrastructure covers LangSmith Deployment, a durable runtime with exactly-once execution, human-in-the-loop approvals, background agents and multi-agent coordination, plus Sandboxes for isolated code execution with filesystem, shell, package-manager and network isolation, and an LLM Gateway that enforces spend caps, PII redaction and routing policy between agents and model providers. On top sits LangSmith Engine, announced in public beta at Interrupt 2026, which autonomously monitors production traces, clusters failures into prioritised issues, diagnoses root causes and opens pull requests proposing fixes. LangSmith Fleet, rebranded from Agent Builder in March 2026 alongside an NVIDIA enterprise partnership, lets non-technical staff describe agents in natural language while administrators retain agent identity management, granular permissions, attribute-based access control and audit logs. The same release cycle shipped SmithDB, a Rust database purpose-built for agent observability that the vendor measures at 92ms P50 trace-tree loads. LangChain raised a $125M Series B led by IVP in October 2025 at a $1.25 billion valuation.
The platform or AI engineering team that owns production agents and is accountable for their reliability — they get one system for tracing, evaluation, deployment and governance instead of stitching four vendors together.
Production agent failures are captured, clustered and diagnosed automatically, so quality regressions surface as prioritised issues with proposed fixes rather than as customer complaints.
At a Glance
- Category
- Agent Development
- Pricing
- Subscription, Usage-based, Freemium, Contact for pricing
- Target Market
- CTOs, AI Engineers, Enterprise Developers, Platform Engineering Leads
- Deployment
- Cloud-first, Hybrid, Self-hosted
- Headquarters
- San Francisco, United States
Key Features
- ✓Trace-level observability
Captures every step of an agent run, including tool calls, model responses and latency, so failures can be reproduced and debugged.
- ✓Production evaluation
Applies LLM-as-judge, code-based and multi-turn evaluators to live production traces, not only to curated offline datasets.
- ✓LangSmith Deployment
Managed durable runtime with exactly-once execution, background agents, human-in-the-loop approvals and multi-agent coordination built in.
- ✓Sandboxes
Generally available isolated environments giving agents a filesystem, shell, package manager and network isolation away from production systems.
- ✓LangSmith Engine
Public-beta agent that clusters production failures into prioritised issues, diagnoses root causes and opens pull requests proposing fixes.
- ✓LLM Gateway
Runtime governance layer between agents and model providers enforcing spend caps, PII redaction and routing policy.
- ✓Fleet
No-code surface letting non-technical teams describe agents in natural language under central identity, permission and audit controls.
Capabilities
Use Cases
- •Debugging a failing support agent
Replay the exact trace of a bad conversation to find which tool call or prompt step broke, then fix and re-evaluate it.
- •Gating agent releases on quality
Run evaluator suites over recorded production traces in CI so a regression blocks the release before any customer sees it.
- •Rolling agents out across business units
Use Fleet with SSO/SAML, ABAC and audit logs to let non-engineers build agents while security retains central control.
- •Controlling model spend
Route every agent model call through the LLM Gateway to enforce per-team spend caps and redact PII before it leaves.
- •Continuous improvement loop
Let Engine cluster recurring production failures into issues and propose pull-request fixes, turning incident triage into reviewed code changes.
Ideal For
Best For
- ✓Debugging multi-step agent runs by inspecting full execution traces down to individual tool calls and model responses
- ✓Running LLM-as-judge and multi-turn evaluations against real production traffic rather than synthetic test sets
- ✓Deploying long-running agents that need durable state, exactly-once execution and human-in-the-loop approval steps
- ✓Governing enterprise agent rollout with SSO/SAML, RBAC/ABAC, SCIM provisioning and organisation-wide audit logs
- ✓Capping and routing model spend across teams through a central LLM gateway with PII redaction
Not Ideal For
- ✗Teams that require self-hosted or air-gapped deployment on a budget — self-hosting and hybrid are reserved for the custom-priced Enterprise tier, a recurring complaint on Hacker News where developers objected to needing a cloud account just to view their own traces
- ✗Small teams wanting predictable flat costs — billing layers $39/seat/month on top of usage-based LangChain Compute Units at $1.50 and Storage Units at $1.00, which is hard to forecast before running real volume
- ✗Organisations standardising on fully open-source observability; LangSmith is proprietary, and several open-source alternatives exist on Hacker News specifically as a reaction to that
Integrations
Deployment
Market Analysis
Pros
- ✓Framework-agnostic — instruments LangGraph, Deep Agents and custom code, and works with closed, open or hybrid model stacks
- ✓Covers the whole agent lifecycle in one product: trace, evaluate, deploy, sandbox, govern spend and auto-diagnose failures
- ✓Strong enterprise governance with SSO/SAML, RBAC/ABAC, SCIM, audit logs and ISO, SOC 2, HIPAA and GDPR compliance
- ✓Backed by a well-capitalised vendor at a $1.25 billion valuation, with an NVIDIA enterprise partnership announced in March 2026
Cons
- ✗Self-hosting is Enterprise-only, and Hacker News threads show developers building open-source alternatives specifically because LangSmith requires a cloud account to view their own traces
- ✗Combined per-seat and usage-based billing makes total cost hard to forecast, and independent comparisons single out the $39/seat model as a drawback
- ✗Proprietary platform whose smoothest experience still favours teams already committed to the LangChain and LangGraph ecosystem
- ✗Several flagship 2026 capabilities — Engine, the LLM Gateway and Managed Deep Agents — remain in beta rather than generally available
Pricing
Developer
$0
- ✓Up to 5k base traces per month, then pay-as-you-go
- ✓Single user
- ✓Community support
Plus
From $39/mo
- ✓Up to 10k base traces per month
- ✓Unlimited additional seats at $39 each
- ✓Access to Deployment, Engine and advanced features
- ✓One free small serverless deployment
Enterprise
Contact for pricing
- ✓Self-hosted and hybrid deployment
- ✓Custom SSO, RBAC and ABAC
- ✓Guaranteed support SLA
- ✓Custom seats and workspace configuration
List pricing is published: Developer is $0 with 5k base traces monthly and Plus is $39 per seat per month with 10k base traces, after which both bill usage-based overage in LangChain Compute Units at $1.50 each and LangChain Storage Units at $1.00 each. That two-axis model makes total cost hard to forecast before production volume is known. Self-hosted and hybrid deployment, custom SSO, RBAC/ABAC and a support SLA are gated behind custom-priced Enterprise, so regulated buyers cannot self-serve.
Security & Compliance
Sources
This page was written from 7 sources, 5 on domains other than langchain.com.
- 1.langchain.com — langsmith platformvendor
- 2.langchain.com — pricing langsmithvendor
- 3.ai2.work — langchain ships langsmith engine and managed deep agents at
- 4.blockchain.news — langchain fleet nvidia partnership enterprise ai agents
- 5.posthog.com — best ai observability tools
- 6.sacra.com — langchain
- 7.hn.algolia.com — hn.algolia.com
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
LangWatch
Open-source agent testing platform pairing simulation-based QA with OpenTelemetry tracing and evals
Composio
Managed tool-calling and authentication layer connecting AI agents to 1,000+ enterprise applications
Kitesurf
An agent-first browser that runs in V8 isolates on Cloudflare Workers
Natural
Payments infrastructure that lets AI agents hold, send and collect money