langsmith

使用LangSmith追踪、评估和部署AI代理及大语言模型应用。适用于添加可观测性、运行评估、设计提示词或…

npx skills add https://github.com/langchain-ai/docs --skill langsmith

LangSmith

LangSmith is a framework-agnostic platform for building, debugging, and deploying AI agents and LLM applications. Trace requests, evaluate outputs, test prompts, and manage deployments all in one place at smith.langchain.com.

When to use

Use LangSmith when you need to:

Trace and debug LLM calls, agent steps, retrieval, and tool use
Evaluate LLM outputs with automated or human-in-the-loop scoring
Engineer prompts with a visual playground and version control
Deploy agents to production with the LangGraph-based agent server
Monitor production systems with dashboards, alerts, and cost tracking

When NOT to use

To build agent logic or LLM pipelines, use LangChain, LangGraph, or Deep Agents instead
LangSmith is the platform layer that complements these frameworks

Quick setup

Set two environment variables to enable tracing from any supported framework:

export LANGSMITH_TRACING=true
export LANGSMITH_API_KEY="your-api-key"  # from smith.langchain.com/settings

Install the SDK

# Python
pip install langsmith

# JavaScript/TypeScript
npm install langsmith

Verify tracing

from langsmith import traceable

@traceable
def my_function(query: str) -> str:
    # Your LLM logic here—all calls inside are traced automatically
    return "result"

Core capabilities

Capability	Description
Observability	Trace every step of your LLM app with automatic or manual instrumentation
Evaluation	Run evaluations with code, LLM-as-judge, or composite evaluators
Prompt engineering	Create, version, and test prompts in a visual playground
Agent deployment	Deploy LangGraph agents with streaming, human-in-the-loop, and durable execution
Monitoring	Dashboards, alerts, and cost tracking for production workloads