What it does
LangSmith is LangChain's observability and evaluation service for applications built with language models and agents. Developers add tracing through supported integrations, SDKs, or custom instrumentation; LangSmith records runs and provides trace inspection, aggregate dashboards, alerts, and online evaluations. Cloud, bring-your-own-cloud, and self-hosted deployments let teams choose who operates the service and where observability data resides.
Follow one request through an agent
A practical first task is to trace a small LLM or agent request. LangSmith's docs describe adding tracing through environment variables, framework integrations, or an SDK, then inspecting the resulting threads and runs in the Messages, Turns, and Details views. A trace makes the sequence of model calls and tool work visible, which helps explain a failure that a final response alone cannot. This is a documented workflow, not a test performed for this record.
Watch trends rather than every trace
Once traces arrive, prebuilt project dashboards summarize measures such as trace volume, latency, errors, token usage, cost, and feedback. Teams can filter and group the data, build custom dashboards, and set alerts or online evaluations. A useful example is tracking latency and error rate for a production support agent, then opening an individual trace when the aggregate chart changes. LangSmith records and organizes telemetry; engineers still need to decide whether a prompt, tool, model, or application change will fix the cause.
LangSmith is not limited to the LangChain framework: its product pages describe SDKs and integrations for other frameworks and custom agent stacks. Its own deployment choices also affect the operating burden. Cloud is managed by LangChain; in bring-your-own-cloud, LangChain manages the control plane while the customer operates data planes; for self-hosting, the customer runs and upgrades the infrastructure. The latter two choices are Enterprise-plan options.
Check trace quotas and hosting terms
The public pricing page lists a free Developer plan for one seat with 5,000 base traces per month, followed by usage charges; Plus is listed at $39 per seat per month with 10,000 base traces, also followed by usage charges. Enterprise pricing is custom. The included amount is not a forecast of a team's complete bill, so estimate trace volume and review current retention and ingestion limits before sending production data. No account was created and no application data was sent during this research.
Strengths
- Trace views expose the runs and threads behind an LLM application request, making it possible to investigate individual failures.
- Prebuilt and custom dashboards summarize trace volume, latency, errors, token use, costs, and feedback scores.
- Teams can configure alerts, online evaluations, and workflow automations around observed traces.
- Cloud, bring-your-own-cloud, and self-hosted deployment choices offer different data-location and infrastructure responsibilities.
Watch out for
- The free and paid plans include trace quotas, then charge for additional usage; the free Developer plan is limited to one seat.
- Self-hosted and bring-your-own-cloud options require the Enterprise plan, and self-hosted customers handle upgrades and scaling.
Best for
- Teams diagnosing errors or unexpected behavior in LLM and agent applications through full execution traces.
- Teams tracking production latency, errors, token use, cost, and evaluation feedback across agent runs.
- Organizations that need to choose between vendor-managed cloud and customer-operated observability infrastructure.
Sources
- Monitor projects with dashboards - Docs by LangChain (Oct 26, 2026)
- LangSmith: Agent & LLM Observability Platform (Oct 26, 2026)
- LangSmith Plans and Pricing (Oct 26, 2026)
- Set up LangSmith - Docs by LangChain (Oct 26, 2026)
- LangSmith Observability - Docs by LangChain (Oct 26, 2026)