LangChain Python Observability Integration

Integrate Spanora with LangChain and LangGraph Python applications using OpenTelemetry auto-instrumentation. Monitor LLM calls, chains, and tools automatically.

Auto-instrument your LangChain / LangGraph Python applications using opentelemetry-instrumentation-langchain. Every LLM call, chain invocation, and tool execution is captured automatically.

No Spanora SDK required — just standard OpenTelemetry.

Installation

pip install langchain langchain-openai \
  opentelemetry-sdk \
  opentelemetry-exporter-otlp \
  opentelemetry-instrumentation-langchain

Setup

main.py

from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.resources import Resource
from opentelemetry.instrumentation.langchain import LangchainInstrumentor

resource = Resource.create({
    "service.name": "my-langchain-app",
    "gen_ai.agent.name": "my-agent",
})

provider = TracerProvider(resource=resource)
provider.add_span_processor(BatchSpanProcessor(OTLPSpanExporter(
    endpoint="https://spanora.ai/api/v1/traces",
    headers={"Authorization": "Bearer <your-spanora-api-key>"},
)))
trace.set_tracer_provider(provider)

# Auto-instrument LangChain (call AFTER setting the provider)
LangchainInstrumentor().instrument()

Basic Usage

Use LangChain as normal — all calls are traced automatically:

chain.py

from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate

llm = ChatOpenAI(model="gpt-4o-mini")
prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful assistant."),
    ("human", "{input}"),
])
chain = prompt | llm

result = chain.invoke({"input": "What is the capital of France?"})

Adding User and Org Context

You can pass context via OpenTelemetry resource attributes or via LangChain's metadata config. Resource attributes are attached to all spans from your service:

resource = Resource.create({
    "service.name": "my-langchain-app",
    "spanora.user.id": "usr_abc123",
    "spanora.org.id": "org_acme",
})

Alternatively, pass metadata per-invocation via LangChain's config. Note that how metadata appears on spans depends on the instrumentor version:

result = chain.invoke(
    {"input": "How do I reset my password?"},
    config={
        "metadata": {
            "user_id": "usr_abc123",
            "org_id": "org_acme",
        }
    },
)

Agent with Tools

agent.py

from langchain_openai import ChatOpenAI
from langchain_core.tools import tool
from langchain.agents import create_agent

@tool
def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    return f"72F and sunny in {city}"

llm = ChatOpenAI(model="gpt-4o")
agent = create_agent(llm, [get_weather])

result = agent.invoke(
    {"messages": [{"role": "user", "content": "What's the weather in Paris?"}]},
    config={"metadata": {"user_id": "usr_abc123", "org_id": "org_acme"}},
)

What Gets Captured

The instrumentor automatically captures:

LLM spans: model, provider, prompts, completions, token counts
Tool spans: tool name, input arguments, output, execution status
Chain spans: full execution hierarchy showing how components are nested
Cost: calculated by Spanora from model + token counts when available
Agent name: set via gen_ai.agent.name resource attribute
User/Org: set via spanora.user.id / spanora.org.id resource attributes

Troubleshooting

Traces not appearing?

Verify your API key is correct and prefixed with Bearer in the auth header
Ensure TracerProvider is set before calling LangchainInstrumentor().instrument()
Check that the endpoint URL ends with /api/v1/traces

Missing prompt content?

The instrumentor captures prompts by default. If content is missing, ensure you haven't set TRACELOOP_TRACE_CONTENT=false.

Missing tokens/cost?

Token counts come from the LLM provider response. Streaming may not report tokens with all providers.
Cost is calculated by Spanora from model + token counts. If the model is not in Spanora's pricing database, cost will be null.

Next Steps

Any OTEL Provider — Send traces from any OTEL-compatible framework or service
OTEL Attributes — Semantic attribute conventions recognized by Spanora

LangChain Python Observability Integration

On this page