LangSmith
Also known as: LangSmith Fleet, SmithDB, LangSmith Engine, Context Hub, LangGraph, LangGraph Platform, LangChain Inc
LangChain's commercial observability, evaluation and agent operations platform, tracing agent runs node by node, running offline and LLM as judge evals, and deploying LangGraph agents as managed durable APIs.
LangSmith is the commercial platform layer built by the LangChain team, and the company's primary revenue driver. The mental model LangChain itself uses is clean: LangChain is the framework, LangGraph is the orchestration runtime, and LangSmith is the platform you point them at to see what happened, measure whether it was good, and run it in production. It entered closed beta in August 2023, reached general availability with paid plans in July 2024, and now handles more than a billion traces. LangChain raised $125 million at a $1.25 billion valuation in October 2025 on the back of it.
The core is tracing and evaluation. Traces carry node by node state diffs, full agent execution graphs and model and tool call breakdowns, and can be replayed against new model versions. Evaluation runs both offline against curated datasets and online through language model as judge evaluators, with prompt versioning and testing alongside. It is framework agnostic on paper, with Python and TypeScript SDKs plus end to end OpenTelemetry support so any compatible application can export to it, but the product is unmistakably LangChain first: instrument a LangGraph agent and the traces, node graph, per step state and deploy path line up with almost no glue code.
Through 2026 it expanded well past observability into an agent operations stack. LangSmith Deployment runs LangGraph agents as managed durable APIs, with Managed Deep Agents added in May 2026. LangSmith Fleet, formerly Agent Builder, handles deployment. LangSmith Engine performs automated failure clustering with proposed fixes, Context Hub versions agent instructions, and an LLM Gateway was added alongside a unified cost view across whole agent workflows. Underneath, SmithDB is a purpose built trace database written in Rust on Apache DataFusion and the Vortex file format, keeping trace data on object storage with stateless ingestion and query.
The trade offs are ecosystem and cost. LangSmith is proprietary: the client SDKs are open source but the backend, interface and storage layer are not, and self hosting is an enterprise only add on, which is the sharpest contrast against Langfuse. Independent reviews are consistent that the cost curve climbs faster than open source alternatives at scale, because trace overages are metered on top of seats.
Vendor details
Canonical URL
https://www.langchain.com/langsmith
Category
Agent infrastructure
Funding status
Private unicorn. 25M USD Series A Feb 2024 led by Sequoia at a reported valuation near 200M USD, the same month LangSmith reached general availability. 125M USD Series B Oct 2025 led by IVP at a reported 1.25B USD valuation, with strategic participation from Databricks, Datadog, Cisco and ServiceNow. Figures are third party reported and were not confirmed against a LangChain newsroom page.
Company status
independent
Use cases & customers
Target customers
Deployment options
In practice
Your LangGraph agent fails in production for reasons traditional monitoring cannot surface. Traces carry node by node state diffs and full execution graphs, and runs can be replayed against new model versions.
You need to prove an agent got better rather than just different. Offline dataset evaluation and online language model as judge evaluators run against the same traced runs.
Failures are piling up faster than anyone can read them. LangSmith Engine clusters failures automatically and proposes fixes rather than leaving a wall of individual traces.
Sources & related URLs
Agentic Index coverage score
9.0 / 14 capabilities · 64%
| Integrations & Tool CallingPython and TypeScript SDKs plus end to end OpenTelemetry support so any compatible application can export traces, zero setup instrumentation of LangChain apps through environment variables alone, an LLM Gateway, and AWS Marketplace availability for enterprise procurement. 2026-08-05 | Full |
|---|---|
| Workflow OrchestrationLangSmith Deployment runs LangGraph agents as managed durable APIs and LangSmith Fleet handles deployment, but authoring the orchestration itself belongs to LangGraph, a separate free framework carried on the langchain record. Graded partial to avoid double counting the framework. 2026-08-05 | Partial |
| Knowledge Grounding & RAGNo retrieval, vector store or knowledge grounding capability. Context Hub versions agent instructions, which is prompt and context management rather than grounding, and retrieval belongs to the application being observed. 2026-08-05 | Unable to verify |
| Human Oversight & GuardrailsHuman review enters through evaluation workflows, curated datasets and prompt testing before promotion, and LangSmith Engine proposes fixes for human acceptance rather than applying them, but no runtime approval gate or guardrail policy over live agent actions is documented. 2026-08-05 | Partial |
| Security, Identity & GovernanceCORRECTED 7 Aug 2026 from a P undergrade. A public trust centre at trust.langchain.com publishes SOC 2 Type II audit reports, penetration test executive summaries, network and system architecture documentation and a subprocessor list with processing locations; LangSmith and LangGraph Platform both hold SOC 2 Type II, and LangSmith documents GDPR and HIPAA compliance with a DPA available. Controls include SSO, role based access control with custom roles, workspaces, data retention policies and PII masking applied before data reaches LangChain servers. LangChain trust centre, security compliance support article and regions FAQ 2026-08-07 | Full |
| Observability & AuditabilityThe core product and the deepest framework level tracing in its field: node by node state diffs, full agent execution graphs, model and tool call breakdowns, replay against new model versions, a unified cost view across whole agent workflows, and LangSmith Engine clustering failures automatically. Over a billion traces handled. 2026-08-05 | Full |
| Memory & State PersistenceSmithDB persists traces on object storage with configurable retention from 14 to 400 days, evaluation datasets and versioned prompts persist, and Context Hub versions agent instructions, but this is platform state rather than agent memory. 2026-08-05 | Partial |
| Deployment & Data ResidencyUPGRADED 7 Aug 2026. Beyond enterprise self hosting, LangSmith documents four selectable regional instances at signup on all plans including free: GCP US, GCP EU, GCP APAC and AWS US, each a separate data residency with identical terms and pricing. Migration between regions is not supported and there is no EU contracting entity, both carried as honest limits. The proprietary backend point still holds against langfuse, but documented multi region residency plus enterprise self hosting clears the axis. LangChain regions FAQ and trust centre 2026-08-07 | Full |
| Prebuilt Agents, Templates & PacksPrebuilt language model as judge evaluators, Managed Deep Agents added May 2026 and Context Hub instruction templates provide ready made components, but there is no catalogue of deployable prebuilt agents or template gallery. 2026-08-05 | Partial |
| Triggers & Channel CoverageDeployed agents are exposed as durable APIs, which is a real invocation surface, but the platform itself is instrumented by applications sending traces rather than offering scheduling, event triggers or channel coverage. 2026-08-05 | Partial |
| Model Flexibility & RoutingAn LLM Gateway was added in the 2026 Interrupt release and runs can be replayed against new model versions, which implies provider portability, but no published model registry, routing policy or fallback configuration was retrieved. 2026-08-05 | Partial |
| APIs, SDKs & MCP ExtensibilityOpen source Python and TypeScript client SDKs, full OpenTelemetry ingestion so any instrumented application can send traces regardless of framework, deployed agents exposed as APIs, and AWS Marketplace distribution. No MCP server documented. 2026-08-05 | Full |
| Testing, Debugging & OptimizationThe strongest evaluation tooling encountered in this sweep: offline evaluation against curated datasets, online language model as judge evaluators, prompt versioning and testing, replay of production runs against new model versions, and automated failure clustering with proposed fixes through LangSmith Engine. 2026-08-05 | Full |
| Browser & Computer UseNo browser control or computer use capability. The platform observes and deploys agents rather than operating interfaces. 2026-08-05 | Unable to verify |
The Agentic Index coverage score grades every vendor Full, Partial or Unable to verify against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
LangSmith launched Managed Deep Agents in public beta. This release allows developers to deploy Deep Agents to a managed LangSmith runtime that includes durable execution, memory, sandboxes, channels, and evaluations.
Bears on: Deployment / data residency
View sourcePricing
Developer free with 5,000 traces per month; Plus $39 per seat per month; Enterprise custom annual
seats plus metered traces
What is public
Tier names, the free trace allowance, the per seat rate and both trace overage rates with their retention windows are all published.
Billing mechanics
Three tiers. Developer is free with 5,000 traces a month. Plus is $39 per seat per month self serve. Enterprise is annual invoiced with custom single sign on, role based access control and self hosted deployment. Above the included quota, traces are metered: $2.50 per 1,000 at 14 day retention or $5.00 per 1,000 at 400 day retention.
Cost watchouts
Trace overages are metered on top of seats and are the dominant production cost, retention tier doubles the per trace rate, and single sign on, role based access control and self hosting are all enterprise only
Variable cost rationale
Cost is driven by trace volume rather than headcount, and trace volume is a function of how much agents do in production, which is precisely the variable teams cannot forecast before deploying. Retention selection doubles the rate again.
Additional watchouts
Two structural costs beyond the sticker. Retention choice doubles the per trace rate, and the governance features most enterprises require, single sign on, role based access control and self hosting, are all gated to the enterprise tier. Independent reviews also flag ecosystem lock in: the cost curve climbs faster than open source alternatives at scale, which is the explicit argument Langfuse makes against it.
Overage / add-ons
Base traces $2.50 per 1,000 with 14 day retention; extended traces $5.00 per 1,000 with 400 day retention, charged once the included quota is exceeded
Sales call required
Mixed (some tiers require a call)
Free / trial
Developer plan free with 5,000 traces per month, enough to instrument and evaluate a real prototype before paying
Lowest paid plan
Plus
Commercial notes
The commercially important fact is that the LangChain and LangGraph frameworks are free and open source while LangSmith is where the bill lands, so a search for LangChain pricing is really a question about LangSmith at production scale. Independent analysis is consistent that trace overages, not seats, drive the invoice, and that costs can reach the thousands per month once agents run in production.
Key ambiguities
The seat rate and overage rates are published, but consumption is expressed through an LCU and LSU usage model that independent analysis describes as hard to forecast before deployment, since cost depends on what agents do in production rather than on plan choice.
Missing data
Enterprise rates, the precise LCU and LSU conversion, and included trace quotas per tier.
Related vendors
- Acrab — Singapore compute infrastructure company building a full stack…
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- AIsa — Unified resource and payment gateway for AI agents that lets them…
- AlphaBitCore — AI control plane that governs how models, agents, tools, and…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
Alternatives to LangSmith
The closest documented capability profiles to LangSmith among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Traceloop9.0 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs
- Arize AI7.5 / 14A lighter documented profile than LangSmith
- Bernstein10.5 / 14Fuller documented coverage on Workflow Orchestration and Triggers & Channel Coverage
- Metorial8.5 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs
- Arcade8.0 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs
- Confident AI7.0 / 14A lighter documented profile than LangSmith
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded