LlamaIndex


Agent structure	Yes — query engine, retriever, synthesizer and LLM spans, nested.
The LLM leg	Yes, with token counts, inside the tree.
Retrieval steps	Yes — this is the reason to use the instrumentor over a base-URL swap.

Agent structure

Yes — query engine, retriever, synthesizer and LLM spans, nested.

The LLM leg

Yes, with token counts, inside the tree.

Retrieval steps

Yes — this is the reason to use the instrumentor over a base-URL swap.

Setup

pip install openinference-instrumentation-llama-index \
            opentelemetry-sdk opentelemetry-exporter-otlp-proto-http

from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.resources import Resource
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from openinference.instrumentation.llama_index import LlamaIndexInstrumentor
import os

provider = TracerProvider(resource=Resource.create({"service.name": "my-rag-app"}))
provider.add_span_processor(BatchSpanProcessor(OTLPSpanExporter(
    endpoint="https://gateway.tracelane.dev/v1/traces",
    headers={"x-tracelane-api-key": os.environ["TRACELANE_API_KEY"]},
)))
trace.set_tracer_provider(provider)

LlamaIndexInstrumentor().instrument(tracer_provider=provider)

Your API key needs the ingest scope.

Span volume

A bare llm.complete() is 1 span. A full query-engine call is many more — retriever, each retrieved node, synthesizer, and the LLM call.

An OTel span and a gateway call are not the same unit, and they are not counted the same way. A gateway request and an OTel span both count as ingest bytes against the SAME meter — there is no separate quota for one path or the other. On this OTel path your app sends spans, and one agent run can emit many spans. Size your ingest estimate against spans per run, not runs, and see pricing.

Verification status

The transport is the standard OTLP/HTTP protobuf exporter and is proven against Tracelane production — the same exporter, endpoint and header carry every Python framework on this page.

The LlamaIndex instrumentor itself was verified locally, not against production. It attaches and emits correctly shaped spans; we have not yet published a production read-back for a full query-engine trace the way we have for LangGraph. If you hit a gap, open an issue.

Get started

SDKs

Integrations

Concepts

Features

Tools

Audit & compliance

Operations

Migrations

Comparisons

What you get

Setup

Span volume

Verification status

​What you get

​Setup

​Span volume

​Verification status

What you get

Setup

Span volume

Verification status