> ## Documentation Index
> Fetch the complete documentation index at: https://arizeai-433a7140.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenLLMetry Tracing

> Trace LLM apps instrumented with Traceloop's OpenLLMetry SDK in Phoenix by converting their OpenTelemetry spans to OpenInference with a span processor.

[OpenLLMetry](https://github.com/traceloop/openllmetry) is Traceloop's open-source set of OpenTelemetry instrumentations for 20+ LLM providers and frameworks. The [Traceloop SDK](https://www.traceloop.com/docs/openllmetry/introduction) (`traceloop-sdk`) is the `Traceloop.init()` wrapper that activates those instrumentations, so this page covers both. The [`openinference-instrumentation-openllmetry`](https://github.com/Arize-ai/openinference/tree/main/python/instrumentation/openinference-instrumentation-openllmetry) span processor converts OpenLLMetry's spans to OpenInference semantic conventions before they're exported to Phoenix.

## Install

```bash theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
pip install traceloop-sdk openinference-instrumentation-openllmetry arize-phoenix-otel openai httpx
```

`traceloop-sdk` imports `httpx` without declaring it as a dependency, so install it explicitly.

## Setup

Set your API keys as environment variables:

```bash theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
export OPENAI_API_KEY=[your_key_here]
export PHOENIX_COLLECTOR_ENDPOINT=http://localhost:6006
export PHOENIX_API_KEY=[your_phoenix_api_key]
```

Pass the OpenInference span processor and a Phoenix exporter to `Traceloop.init()`. The conversion processor must come before the exporting processor:

```python instrumentation.py theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
from openinference.instrumentation.openllmetry import OpenInferenceSpanProcessor
from phoenix.otel import PROJECT_NAME, HTTPSpanExporter, SimpleSpanProcessor
from traceloop.sdk import Traceloop

project_name = "openllmetry"

# HTTPSpanExporter reads PHOENIX_COLLECTOR_ENDPOINT and PHOENIX_API_KEY.
Traceloop.init(
    app_name=project_name,
    disable_batch=True,
    processor=[OpenInferenceSpanProcessor(), SimpleSpanProcessor(HTTPSpanExporter())],
    resource_attributes={PROJECT_NAME: project_name},
)
```

OpenLLMetry instruments every provider it supports the same way. To trace a different provider, swap the OpenAI client below for [another supported provider](https://www.traceloop.com/docs/openllmetry/tracing/supported) and keep the setup unchanged.

> [!NOTE]
>
> Prefer individual instrumentors to the Traceloop SDK? Install the specific OpenLLMetry instrumentor, such as `opentelemetry-instrumentation-openai`, and instrument it against a Phoenix tracer provider that has `OpenInferenceSpanProcessor()` added before the exporter. The conversion step is the same.

## Run OpenLLMetry

```python example.py theme={"theme":{"light":"github-light-default","dark":"github-dark-default"}}
# Import instrumentation first so Traceloop.init() activates OpenLLMetry's
# instrumentation before the OpenAI client is used.
from instrumentation import project_name  # noqa: F401

from openai import OpenAI

# The client reads OPENAI_API_KEY from the environment.
client = OpenAI()

response = client.chat.completions.create(
    model="gpt-5.4-mini",
    messages=[
        {"role": "user", "content": "Write a haiku about observability."},
    ],
)

print(response.choices[0].message.content)
```

## Observe

Open the `openllmetry` project in Phoenix. The trace contains an `openai.chat` span with its span kind set to LLM, the prompt and response, the model name, and token usage.

## Resources

* [OpenInference OpenLLMetry span processor](https://github.com/Arize-ai/openinference/tree/main/python/instrumentation/openinference-instrumentation-openllmetry)
* [OpenLLMetry on GitHub](https://github.com/traceloop/openllmetry)
* [Traceloop SDK documentation](https://www.traceloop.com/docs/openllmetry/introduction)