langchain-observability

# LangChain Observability ## Overview Production observability for LangChain: LangSmith tracing (zero-code), custom Prometheus metrics, OpenTelemetry integration, structured logging, and Grafana dashboards. ## Tier 1: LangSmith Tracing (Zero-Code Setup) LangSmith automatically traces all LangChain calls when env vars are set. ```bash # Add to .env — that's it. No code changes needed. LANGSMITH_TRACING=true LANGSMITH_API_KEY=lsv2_pt_... LANGSMITH_PROJECT=my-app-production # Optional: background callbacks for lower latency (non-serverless) LANGCHAIN_CALLBACKS_BACKGROUND=true ``` Every chain, LLM call, tool invocation, and retriever query is automatically traced with: - Input/output payloads - Token usage and cost - Latency per step - Error details with stack traces - Parent-child run relationships ### Query Traces Programmatically ```typescript import { Client } from "langsmith"; const client = new Client(); // Get recent failed runs const failedRuns = client.listRuns({ projectName: "my-app-production", error: true, limit: 10, }); for await (const run of failedRuns) { console.log(`${run.name}: ${run.error} (${run.totalTokens} tokens)`); } ``` ## Tier 2: Custom Metrics Callback ```typescript import { BaseCallbackHandler } from "@langchain/core/callbacks/base"; interface Metrics { totalRequests: number; totalErrors: number; totalTokens: number; latencies: number[]; } class MetricsCallback extends BaseCallbackHandler { name = "MetricsCallback"; ...