Skip to main content
dd-trace instruments AI/ML SDKs to automatically capture model inputs, outputs, token counts, and latency. These spans feed into LLM Observability in the Datadog UI.

​
Supported integrations

OpenAI

openai

Anthropic

@anthropic-ai/sdk

LangChain

langchain / @langchain/core

LangGraph

@langchain/langgraph

Vertex AI

@google-cloud/vertexai

Google GenAI

@google/genai

AWS Bedrock

@aws-sdk/smithy-client

Vercel AI SDK

ai

​
Enabling LLM Observability

LLM Observability is automatically active when the AI plugins are loaded. To send LLM spans to the Datadog LLM Observability product, set the following environment variables: 
DD_LLMOBS_ENABLED=true
DD_LLMOBS_ML_APP=my-llm-app
DD_API_KEY=<your-api-key>
Or configure programmatically: 
const tracer = require('dd-trace').init({
  llmobs: {
    mlApp: 'my-llm-app'
  }
})
For OpenAI metrics (token usage, request counts) to work, DogStatsD must be enabled in the Datadog Agent.

​
OpenAI

const tracer = require('dd-trace').init()

tracer.use('openai', {
  service: 'openai'
})

const OpenAI = require('openai')
const client = new OpenAI()

// Spans are created automatically for completions, embeddings, etc.
const response = await client.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: 'Hello' }]
})
The plugin captures:
  • Model name and version
  • Input prompt and completion text
  • Token counts (prompt, completion, total)
  • Request duration and status
Prompt and completion content may contain sensitive data. Review your data retention settings in Datadog before enabling content capture in production.

​
Anthropic

const tracer = require('dd-trace').init()

tracer.use('anthropic', {
  service: 'anthropic'
})

const Anthropic = require('@anthropic-ai/sdk')
const client = new Anthropic()

// Automatically traced
const message = await client.messages.create({
  model: 'claude-opus-4-5',
  max_tokens: 1024,
  messages: [{ role: 'user', content: 'Hello' }]
})

​
LangChain

const tracer = require('dd-trace').init()

tracer.use('langchain', {
  service: 'langchain'
})

// Works with @langchain/core and @langchain/openai
const { ChatOpenAI } = require('@langchain/openai')
const model = new ChatOpenAI({ model: 'gpt-4o' })

// Chain invocations, tool calls, and retriever calls are all traced
const result = await model.invoke('What is the capital of France?')

​
LangGraph

const tracer = require('dd-trace').init()

tracer.use('langgraph', {
  service: 'langgraph'
})

// Graph executions, node transitions, and state updates are traced
const { StateGraph } = require('@langchain/langgraph')

​
Google Cloud Vertex AI

const tracer = require('dd-trace').init()

tracer.use('google-cloud-vertexai', {
  service: 'vertexai'
})

const { VertexAI } = require('@google-cloud/vertexai')
const vertexai = new VertexAI({ project: 'my-project', location: 'us-central1' })

// Automatically traced
const model = vertexai.getGenerativeModel({ model: 'gemini-2.0-flash' })
const result = await model.generateContent('Hello')

​
Google GenAI

const tracer = require('dd-trace').init()

tracer.use('google-genai', {
  service: 'google-genai'
})

const { GoogleGenerativeAI } = require('@google/genai')
const genai = new GoogleGenerativeAI(process.env.GOOGLE_API_KEY)

// Automatically traced
const model = genai.getGenerativeModel({ model: 'gemini-2.0-flash' })
const result = await model.generateContent('Hello')

​
AWS Bedrock (via aws-sdk)

Bedrock is instrumented through the aws-sdk plugin: 
const tracer = require('dd-trace').init()

tracer.use('aws-sdk', {
  service: 'bedrock'
})

// @aws-sdk/client-bedrock-runtime is automatically traced
const { BedrockRuntimeClient, InvokeModelCommand } = require('@aws-sdk/client-bedrock-runtime')
const client = new BedrockRuntimeClient({ region: 'us-east-1' })

const response = await client.send(new InvokeModelCommand({
  modelId: 'anthropic.claude-opus-4-5',
  body: JSON.stringify({ prompt: 'Hello', max_tokens_to_sample: 256 })
}))

​
Vercel AI SDK

const tracer = require('dd-trace').init()

tracer.use('ai', {
  service: 'vercel-ai'
})

// Automatically traced when using the `ai` package
const { generateText } = require('ai')
const { openai } = require('@ai-sdk/openai')

const { text } = await generateText({
  model: openai('gpt-4o'),
  prompt: 'Hello'
})

Build docs developers (and LLMs) love

Get started for freeTalk to us