Master LangSmith with our step-by-step tutorial, detailed feature walkthrough, and expert tips.
Sign up at smith.langchain.com and create a new project for your LLM application Set LANGCHAIN_TRACING_V2=true and LANGCHAIN_API_KEY environment variables to enable automatic tracing For LangChain apps, traces appear automatically; for other frameworks, use the LangSmith SDK or OpenTelemetry integration Create an evaluation dataset with example inputs and reference outputs, then run your first evaluation experiment Set up production monitoring dashboards to track latency, error rates, and token costs across all LLM operations
💡 Quick Start: Follow these 1 steps in order to get up and running with LangSmith quickly.
Explore the key features that make LangSmith powerful for ai observability workflows.
No, LangSmith works with any LLM application through its Python/TypeScript SDK or OpenTelemetry integration. You can instrument custom code, direct API calls to OpenAI/Anthropic, or applications built with other frameworks like LlamaIndex or Haystack. However, LangChain and LangGraph applications get the best experience with near-zero-configuration tracing — just a few environment variables enable full capture. If you don't use LangChain at all, alternatives like Langfuse or Helicone may offer a more framework-neutral experience with comparable feature sets.
You create datasets of example inputs (and optionally reference outputs), define evaluator functions that score your application's outputs, and run evaluation experiments against those datasets. Evaluators can be LLM-based (using a judge model like GPT-4 to grade quality), heuristic (regex, string matching, JSON validation, exact match), or human (manual review in the UI by annotators). LangSmith tracks results over time and lets you compare runs across different prompts, models, or retrieval strategies in side-by-side views. This evaluation-first workflow is critical for catching regressions when changing prompts, models, or retrieval pipelines before they reach production users.
LangSmith's free Developer tier includes 5,000 traces/month, sufficient for development but not production-scale traffic. The Plus tier starts at $39 per user per month and includes 10,000 base traces, with additional traces at $0.50 per 1,000 and extended retention available as an add-on. Enterprise pricing is custom with unlimited traces, SSO, RBAC, audit logs, and dedicated support typically sold on annual contracts. For high-volume production applications generating millions of traces monthly, costs can reach four or five figures — this is where self-hosted alternatives like Langfuse become significantly more cost-effective.
LangSmith is primarily a closed-source, hosted SaaS platform with US and EU cloud regions available. Self-hosted deployment is only offered as part of Enterprise contracts and requires direct sales engagement — it is not available on Plus or Developer tiers. This is a significant limitation for enterprises with strict data residency requirements or those who prefer to keep all LLM inputs and outputs within their own infrastructure. LangSmith does offer SOC 2 Type II compliance and data processing agreements, but organizations requiring fully open self-hosting at lower price points should consider Langfuse, Helicone, or Arize Phoenix.
LangSmith and Langfuse cover similar feature surfaces — tracing, evaluation, prompt management, and dashboards — but differ on licensing and ecosystem fit. LangSmith is closed-source, hosted by LangChain Inc., and offers first-class integration with the LangChain/LangGraph framework with auto-instrumentation. Langfuse is open-source (MIT licensed), can be self-hosted for free at any scale, and is framework-neutral with strong SDKs for Python, TypeScript, and Java. Choose LangSmith if you live in the LangChain ecosystem and value polish; choose Langfuse if you need self-hosting, predictable costs at high volume, or framework independence.
Now that you know how to use LangSmith, it's time to put this knowledge into practice.
Sign up and follow the tutorial steps
Check pros, cons, and user feedback
See how it stacks against alternatives
Follow our tutorial and master this powerful ai observability tool in minutes.
Tutorial updated March 2026