@fiddler-ai/langgraph

LangGraph JS instrumentation for Fiddler observability. Auto-traces LangGraph applications via OpenTelemetry using the @fiddler-ai/otel SDK.

Installation

npm install @fiddler-ai/langgraph @fiddler-ai/otel @opentelemetry/api @langchain/core

Quick Start

import { FiddlerClient } from "@fiddler-ai/otel";
import { LangGraphInstrumentor, setConversationId } from "@fiddler-ai/langgraph";

// 1. Create a Fiddler client
const client = new FiddlerClient({
  url: "https://app.fiddler.ai",
  apiKey: process.env.FIDDLER_API_KEY!,
  applicationId: "your-app-uuid",
});

// 2. Auto-instrument LangGraph
const instrumentor = new LangGraphInstrumentor(client);
await instrumentor.instrument();

// 3. Set conversation context
setConversationId("session-123");

// 4. Use LangGraph normally — traces are auto-captured
const result = await graph.invoke({ messages: [new HumanMessage("Hello")] });

// 5. Flush before exit
await client.flush();

Manual Callback Handler

For more control, use the callback handler directly:

import { FiddlerClient } from "@fiddler-ai/otel";
import { FiddlerCallbackHandler } from "@fiddler-ai/langgraph";

const client = new FiddlerClient({ /* ... */ });
const handler = new FiddlerCallbackHandler(client);

const result = await graph.invoke(
  { messages: [new HumanMessage("Hello")] },
  { callbacks: [handler] },
);

Context Utilities

import {
  setConversationId,
  addSessionAttributes,
  setLlmContext,
  runWithContext,
} from "@fiddler-ai/langgraph";

// Group traces by conversation
setConversationId("thread-abc");

// Add user metadata to all spans
addSessionAttributes({ userId: "user-123", environment: "production" });

// Attach RAG context to a model
setLlmContext(chatModel, "Retrieved context: ...");

// Isolate context in async boundaries
runWithContext(() => {
  setConversationId("isolated-session");
  // spans here get "isolated-session"
});

Requirements

• Node.js >= 20
• @langchain/core >= 0.3.0
• @fiddler-ai/otel >= 0.1.0
• @opentelemetry/api >= 1.9.0