LangSmith
FreemiumObserve, evaluate, and deploy reliable AI agents.
Published 1 October 2026
Scores
Popularity3/5
The default observability choice for LangChain users and the brand most often named first in LLM tracing comparisons, but the category is still young and little known outside teams building with LLMs.
Learning Curve2/5
LangChain and LangGraph apps start tracing by setting an API key and one environment variable, and the UI is built around the concepts those frameworks already use. Evaluations and deployment add more to learn.
Flexibility4/5
Covers tracing, evals, prompts, monitoring, and deployment with SDKs and OpenTelemetry for any framework, but customisation stops at what the hosted product and its API expose.
Performance4/5
Handles production trace volumes for large LangChain customers, and its managed deployment runtime supports persistence and background runs for long agents.
Portability2/5
Traces and datasets live in LangChain's cloud unless an Enterprise contract pays for self-hosting, and agents on LangSmith Deployment are tied to its runtime, although OpenTelemetry ingestion eases instrumenting outside LangChain.
About LangSmith
LangSmith is the agent engineering platform from LangChain, the company behind the LangChain and LangGraph frameworks. It started as a tracing and debugging tool for LangChain apps and has grown into a hosted stack for observing, evaluating, and running agents in production. It is framework-agnostic: SDKs for Python and TypeScript, a traceable decorator for plain code, and end-to-end OpenTelemetry support let any application send traces, though the deepest views are built around LangGraph.
Its core is tracing: every request becomes a trace of nested runs covering model calls, tool calls, retrieval, and agent steps, with token, cost, and latency data attached. Monitoring dashboards and alerts sit on top, and the Insights Agent clusters production traces to surface common usage patterns and failure modes. Evaluations cover offline test datasets, online scoring of live traffic, LLM-as-judge and heuristic evaluators, pairwise comparisons, multi-turn conversation evals, and human annotation queues. A prompt hub and playground handle prompt versioning and testing.
Beyond observability, LangSmith now covers agent deployment. LangSmith Deployment (formerly LangGraph Platform) hosts agents as scalable APIs with persistence and background runs, Studio provides a visual debugger for stepping through graphs, and Fleet (formerly Agent Builder) lets non-technical staff describe agents in natural language under admin-controlled permissions.
LangSmith is a proprietary, cloud-first service with a free Developer plan, a per-seat Plus plan billed with trace overages, and an Enterprise plan that adds self-hosting in the customer's own cloud, SSO, and compliance features. Teams that want an open-source, freely self-hostable equivalent usually choose Langfuse instead.
Key Features
- Nested tracing of model calls, tools, retrieval, and agent steps
- Monitoring dashboards and alerts with cost and latency tracking
- Offline and online evaluations with LLM-as-judge, heuristic, and human reviewers
- Insights Agent that clusters traces into usage patterns and failure modes
- Prompt hub and playground for prompt versioning
- Studio visual debugger for stepping through LangGraph agents
- Managed agent hosting through LangSmith Deployment
- OpenTelemetry ingestion for apps outside the LangChain ecosystem
Pros
- Zero-config tracing for LangChain and LangGraph apps through one environment variable
- The most complete evaluation tooling in the category, including multi-turn and pairwise evals
- Observability, evals, and deployment in one product for teams already on LangGraph
- Free Developer plan is enough for individual prototyping
Cons
- Self-hosting is limited to the Enterprise plan
- Per-seat pricing plus trace overages gets expensive for larger teams
- Closed-source, so the platform itself cannot be audited or extended
- Non-LangChain stacks work through SDKs or OTel but miss some LangGraph-specific views
LangSmith Pricing
Freemium- · One seat with a monthly allowance of free traces
- · Tracing, evals, and prompt playground
- · Per seat, with 10,000 base traces included
- · $2.50 per 1,000 extra base traces with 14-day retention
- · One free development deployment
- · Self-hosting in your own cloud, SSO, and compliance features
- · Custom pricing from sales
Tech Stacks with LangSmith
Python AI Agent
ProjectA code-first AI agent in Python. An agent framework such as LangGraph or Pydantic AI runs the loop, FastAPI serves it, PostgreSQL (your own, Neon, or Supabase) stores its state, and a model such as Claude or OpenAI does the reasoning.
TypeScript AI Agent
ProjectA code-first AI agent in TypeScript. Next.js holds the agent and its chat interface in one app, a framework such as the Vercel AI SDK or Mastra runs the loop, PostgreSQL (your own, Neon, or Supabase) keeps the conversations, and a model such as Claude or OpenAI does the reasoning.
Programming
Databases
Hosting
LLM
Agentic
Python RAG App
ProjectA retrieval-augmented generation (RAG) app in Python. A framework such as LlamaIndex or LangChain loads, splits, and indexes your documents, a vector store such as PostgreSQL with pgvector or Qdrant finds the passages that match a question, FastAPI serves the answers, and a model such as Claude or OpenAI writes them from what was retrieved.
Tools Related to LangSmith
Integrates with LangSmith(2)
LangSmith is built by the LangChain team, and LangChain apps send traces to it by setting an environment variable, with no code changes.
LangSmith traces LangGraph agents natively, steps through their graphs in Studio, and hosts them as APIs through LangSmith Deployment, formerly LangGraph Platform.
Alternatives to LangSmith(1)
Langfuse is open source and free to self-host, with a framework-neutral design; LangSmith is proprietary and cloud-first, with the deepest debugging views for LangChain and LangGraph and managed agent deployment on top.
Vendor
LangChain
Website →Tags
Details
- Maintained
- Yes
- Monitoring
- Llm Tracing, Evaluations, Metrics
- Hosting
- Cloud & Self-hosted
- Open source
- No
- Alerting
- Yes