LangChain / LangGraph

About the LangChain integration

Instrument LangChain with any OpenTelemetry instrumentation — for example OpenLLMetry — and point the OTLP exporter at our endpoint. Chains, model calls, and tool steps arrive as spans. telemetry.dev detects LangChain and normalizes its span shapes, including traceloop.* attributes, into one model of spans, tokens, and cost.

Key features

• Works with your existing OTel setup: Keep the instrumentation you already use and send standard OTLP/HTTP.
• Chains, tools, and model calls in one trace: Follow a run from retrieval to tool use to model response.
• Server-side cost from real token usage: Cost is computed from emitted token counts and current model pricing.
• Group by model, provider, environment, and user: Break spend and latency down by the dimensions your app emits.
• Standard OTLP, no lock-in: Your traces stay portable because the transport is OpenTelemetry.

Get started

Python with OpenLLMetry

This example uses OpenLLMetry and LangChain's ChatOpenAI integration. It sends one real OpenAI request. OpenAI charges apply.

Use Python 3.10 or later. Install the packages in a virtual environment:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install traceloop-sdk langchain langchain-openai

Get a project API key from telemetry.dev. Set the keys and the OpenLLMetry exporter configuration:

export TELEMETRY_DEV_API_KEY="your-project-api-key"
export OPENAI_API_KEY="your-openai-api-key"
export TRACELOOP_BASE_URL="https://ingest.telemetry.dev"
export TRACELOOP_HEADERS="Authorization=Bearer%20$TELEMETRY_DEV_API_KEY"

Keep these keys out of version control. A Traceloop Cloud account is not necessary. OpenLLMetry sends traces to https://ingest.telemetry.dev/v1/traces.

Save this code as first_trace.py. Initialize OpenLLMetry before you construct the chain:

from traceloop.sdk import Traceloop
from traceloop.sdk.instruments import Instruments

Traceloop.init(
    app_name="langchain-first-trace",
    resource_attributes={"deployment.environment.name": "development"},
    instruments={Instruments.LANGCHAIN},
    disable_batch=True,
    telemetry_enabled=False,
)

from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI

prompt = ChatPromptTemplate.from_messages([("human", "{question}")])
chain = prompt | ChatOpenAI(model="gpt-4o-mini", max_tokens=100)
response = chain.invoke(
    {"question": "What is OTLP?"},
    config={"run_name": "first-chain"},
)
print(response.content)

Run the file:

python first_trace.py

Open your telemetry.dev project and select the development environment. Find the trace from langchain-first-trace and open its chain and model spans. A successful model call records token usage and latency. The server computes cost from the reported tokens and model pricing.

disable_batch=True sends each span when the span ends, so this short script does not wait for a batch interval. For other exporter options, read the OpenLLMetry configuration guide.

Existing OpenTelemetry setup

These variables configure an exporter. They do not install or initialize LangChain instrumentation. For a new Python app, use the complete setup above.

For an app that already emits LangChain spans, configure its OTLP/HTTP exporter:

export OTEL_EXPORTER_OTLP_ENDPOINT="https://ingest.telemetry.dev"
export OTEL_EXPORTER_OTLP_PROTOCOL="http/protobuf"
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer%20$TELEMETRY_DEV_API_KEY"
export OTEL_SERVICE_NAME="your-service-name"
export OTEL_RESOURCE_ATTRIBUTES="deployment.environment.name=production"

Set TELEMETRY_DEV_API_KEY to your project API key before you run these commands. The exporter sends traces to https://ingest.telemetry.dev/v1/traces.

Prefer tracing the model calls directly?

Because LangChain calls the provider SDKs underneath, our provider wrappers work inside chains too: instrument the openai, anthropic, or google-genai client with telemetry-dev-openai, telemetry-dev-anthropic, or telemetry-dev-google-genai and every model call your chain makes is traced with tokens and cost — no OpenLLMetry required.