MLflow Via OpenInference

Send OpenInference spans from an async YFinance agent to an MLflow tracking server over the OTLP HTTP trace endpoint.

Demonstrates instrumenting an Agno agent with OpenInference and sending traces to MLflow.

mlflow_via_openinference.py
"""
MLflow Via OpenInference
========================

Demonstrates instrumenting an Agno agent with OpenInference and sending traces to MLflow.

Requirements:
    pip install -U mlflow opentelemetry-exporter-otlp-proto-http openinference-instrumentation-agno

Start MLflow with OTLP tracing enabled:
    mlflow server --host 127.0.0.1 --port 5000
"""

import asyncio
import os

from agno.agent import Agent
from agno.models.openai import OpenAIChat
from agno.tools.yfinance import YFinanceTools
from openinference.instrumentation.agno import AgnoInstrumentor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor

# ---------------------------------------------------------------------------
# Setup
# ---------------------------------------------------------------------------
MLFLOW_TRACKING_URI = os.getenv("MLFLOW_TRACKING_URI", "http://127.0.0.1:5000")

endpoint = f"{MLFLOW_TRACKING_URI}/api/2.0/mlflow/traces"

tracer_provider = TracerProvider()
tracer_provider.add_span_processor(
    SimpleSpanProcessor(
        OTLPSpanExporter(
            endpoint=endpoint,
            headers={"x-mlflow-experiment-id": "0"},
        )
    )
)
# Start instrumenting agno
AgnoInstrumentor().instrument(tracer_provider=tracer_provider)


# ---------------------------------------------------------------------------
# Create Agent
# ---------------------------------------------------------------------------
agent = Agent(
    name="Stock Price Agent",
    model=OpenAIChat(id="gpt-5.6-luna"),
    tools=[YFinanceTools()],
    instructions="You are a stock price agent. Answer questions in the style of a stock analyst.",
)


# ---------------------------------------------------------------------------
# Run Example
# ---------------------------------------------------------------------------
async def main() -> None:
    await agent.aprint_response(
        "What is the current price of Tesla? Then find the current price of NVIDIA",
        stream=True,
    )


if __name__ == "__main__":
    asyncio.run(main())

Current MLflow OTLP endpoint

Use MLflow 3.6 or later with a SQL-backed tracking store. Before constructing OTLPSpanExporter, replace the source's REST endpoint assignment with the OTLP endpoint:

endpoint = f"{MLFLOW_TRACKING_URI.rstrip('/')}/v1/traces"

The header selects experiment 0; replace it with your experiment's ID if needed. See MLflow OpenTelemetry ingestion.

Run the Example

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate

Install dependencies

uv pip install -U agno "mlflow>=3.6" openai openinference-instrumentation-agno opentelemetry-exporter-otlp opentelemetry-sdk yfinance

Export your OpenAI API key

export OPENAI_API_KEY="your_openai_api_key_here"

Start MLflow

Start the local MLflow receiver on port 5000 and leave it running in a separate terminal:

mlflow server --host 127.0.0.1 --port 5000

Run the example

Save the code above as mlflow_via_openinference.py, then run:

python mlflow_via_openinference.py

Full source: cookbook/observability/mlflow_via_openinference.py