Microsoft Agent Framework Tracing for .NET

Microsoft Agent Framework is Microsoft’s open-source SDK for building AI agents. Its C# SDK emits OpenTelemetry spans that follow the OpenTelemetry GenAI semantic conventions. Phoenix converts those spans to OpenInference when they arrive, so you only need the OpenTelemetry SDK and an OTLP exporter. No OpenInference package is required. See Translating semantic conventions.

Prerequisites

.NET 10 SDK
An OPENAI_API_KEY from the OpenAI Platform

Launch Phoenix

Local
Container
Self-Host

uvx arize-phoenix serve

No uv? pip install arize-phoenix && phoenix serve does the same thing. See Terminal setup for customization.

docker run -p 6006:6006 -p 4317:4317 arizephoenix/phoenix:latest

Images are published to Docker Hub. See Docker for volumes, PostgreSQL, and other options.

Phoenix serves its UI and OTLP HTTP on port 6006, and OTLP gRPC on port 4317. For a local instance that’s http://localhost:6006. Leave it running while you work.

Install

Create a console project and add the Microsoft Agent Framework OpenAI connector and the OpenTelemetry Protocol exporter:

dotnet new console --framework net10.0 --name MicrosoftAgentFramework
cd MicrosoftAgentFramework
dotnet add package Microsoft.Agents.AI.OpenAI --version 1.24.0
dotnet add package OpenTelemetry.Exporter.OpenTelemetryProtocol --version 1.19.1

Setup

Set the Phoenix endpoint, the project name, and your OpenAI key:

export PHOENIX_COLLECTOR_ENDPOINT="http://localhost:6006"
export PHOENIX_PROJECT_NAME="microsoft-agent-framework-csharp"
export OPENAI_API_KEY="<your-openai-api-key>"

# Only if your Phoenix instance has authentication enabled
# export PHOENIX_API_KEY="<your-phoenix-api-key>"

Replace the contents of Program.cs with the code below. It configures an OTLP HTTP exporter that sends spans to <PHOENIX_COLLECTOR_ENDPOINT>/v1/traces, sets the Phoenix project with the openinference.project.name resource attribute, and builds one agent with a GetWeather function tool. GPT-6.1 Sol requires the Responses API for tool calling, so the agent uses GetResponsesClient().

EnableSensitiveData = true records prompts, responses, tool arguments, and tool results on the spans. Enable it only when you trust the trace destination. Set it to false to omit message content.

Program.cs

﻿#pragma warning disable OPENAI001

using Microsoft.Agents.AI;
using Microsoft.Extensions.AI;
using OpenAI;
using OpenAI.Responses;
using OpenTelemetry;
using OpenTelemetry.Exporter;
using OpenTelemetry.Resources;
using OpenTelemetry.Trace;

const string sourceName = "Microsoft.AgentFramework.Sample";
var phoenixEndpoint = Environment.GetEnvironmentVariable("PHOENIX_COLLECTOR_ENDPOINT")
    ?? "http://localhost:6006";
var phoenixApiKey = Environment.GetEnvironmentVariable("PHOENIX_API_KEY");
var projectName = Environment.GetEnvironmentVariable("PHOENIX_PROJECT_NAME")
    ?? "microsoft-agent-framework-csharp";
var openAiKey = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;

using var tracerProvider = Sdk.CreateTracerProviderBuilder()
    .SetResourceBuilder(ResourceBuilder.CreateDefault()
        .AddService("microsoft-agent-framework-csharp")
        .AddAttributes(new Dictionary<string, object>
        {
            ["openinference.project.name"] = projectName
        }))
    .AddSource(sourceName)
    .AddOtlpExporter(options =>
    {
        options.Endpoint = new Uri($"{phoenixEndpoint.TrimEnd('/')}/v1/traces");
        options.Protocol = OtlpExportProtocol.HttpProtobuf;
        if (!string.IsNullOrEmpty(phoenixApiKey))
        {
            options.Headers = $"Authorization=Bearer {phoenixApiKey}";
        }
    })
    .Build();

Console.WriteLine("Phoenix tracing initialized.");

var openAiClient = new OpenAIClient(openAiKey);

var agent = openAiClient.GetResponsesClient().AsAIAgent(
        model: "gpt-6.1-sol",
        name: "weather-agent",
        instructions:
            "You are a concise weather assistant. Always call the GetWeather " +
            "tool once before answering.",
        tools: [AIFunctionFactory.Create(WeatherTools.GetWeather)])
    .AsBuilder()
    .UseOpenTelemetry(
        sourceName, telemetry => telemetry.EnableSensitiveData = true)
    .Build();

var result = await agent.RunAsync("What is the weather in San Francisco?");
Console.WriteLine(result);

UseOpenTelemetry instruments the agent and emits spans for the agent invocation, each chat call, and each function execution. The tracer provider listens to the same source name passed to UseOpenTelemetry and exports those spans to Phoenix.

Add a tool

Create WeatherTools.cs for the function tool the agent calls:

WeatherTools.cs

using System.ComponentModel;

static class WeatherTools
{
    [Description("Returns the current weather for a city.")]
    public static string GetWeather(
        [Description("The city to check.")] string city) =>
        $"Sunny and 72°F in {city}.";
}

Run Microsoft Agent Framework

Run the project from the MicrosoftAgentFramework directory:

dotnet run

You should see output like:

Phoenix tracing initialized.
It's sunny and 72°F in San Francisco.

Observe

Open Phoenix and select the microsoft-agent-framework-csharp project. Each agent.RunAsync() call appears as one trace:

Span	Span kind	What it shows
`invoke_agent weather-agent(...)`	`AGENT`	The root span for the run, with the agent name, the final response, the model, and total token usage
`chat gpt-6.1-sol`	`LLM`	Each model call, with input and output messages, the model name, and token counts. The agent makes one call to request the tool and one to write the answer
`execute_tool GetWeather`	`TOOL`	The function call, with its arguments and result

Troubleshooting

No traces in Phoenix. Confirm PHOENIX_COLLECTOR_ENDPOINT points at your Phoenix instance and is set in the same shell that runs dotnet run. The exporter appends /v1/traces to it. If authentication is enabled, set PHOENIX_API_KEY.
No agent spans. The source name passed to UseOpenTelemetry must match the source registered with .AddSource(sourceName).
Spans land in the default project. Confirm the openinference.project.name resource attribute is set.
Prompts or tool content are missing. Set EnableSensitiveData = true in the agent’s OpenTelemetry configuration.
401 from OpenAI. Verify OPENAI_API_KEY is valid and can access gpt-6.1-sol. Replace the model with one available to your account if needed.

Developer Tools

LLM Providers

TypeScript

Python

Java

.NET

Platform

Vector Databases

Sandboxes

Evaluation Integrations

Coding Agents

Microsoft Agent Framework Tracing for .NET

Prerequisites

Launch Phoenix

Install

Setup

Add a tool

Run Microsoft Agent Framework

Observe

Troubleshooting

Resources

Microsoft Agent Framework documentation

Microsoft Agent Framework observability

Microsoft Agent Framework on GitHub

​Prerequisites

​Launch Phoenix

​Install

​Setup

​Add a tool

​Run Microsoft Agent Framework

​Observe

​Troubleshooting

​Resources

Microsoft Agent Framework documentation

Microsoft Agent Framework observability

Microsoft Agent Framework on GitHub

Prerequisites

Launch Phoenix

Install

Setup

Add a tool

Run Microsoft Agent Framework

Observe

Troubleshooting

Resources