How to trace Semantic Kernel with HoneyHive

Semantic Kernel is Microsoft’s open-source SDK for building AI agents and multi-agent systems. It provides a model-agnostic framework with plugins, function calling, and enterprise-ready orchestration. HoneyHive works best with Semantic Kernel when you combine SK’s native OpenTelemetry diagnostics with a provider-specific instrumentor. SK-native spans preserve agent names and orchestration structure, while the provider instrumentor adds richer model input/output payloads.

Quick Start

Recommended setup. Enable SK’s native diagnostics before imports, initialize HoneyHive, then instrument your model provider so model events include full chat history and responses.

The examples on this page use OpenAIChatCompletion, so they use OpenAIInstrumentor. If your Semantic Kernel app uses Anthropic, Bedrock, Gemini, or another provider, use the matching provider openinference instrumentor when one exists.

To see where to initialize the tracer for your environment, including AWS Lambda and long-running servers, see Tracer Initialization.

Last tested with semantic-kernel 1.36.0 (May 2026).

pip install honeyhive semantic-kernel openinference-instrumentation-openai

import os

# 1. Enable SK's native OpenTelemetry (before any other imports)
os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS"] = "true"
os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE"] = "true"

# 2. Initialize HoneyHive
from honeyhive import HoneyHiveTracer
from openinference.instrumentation.openai import OpenAIInstrumentor

tracer = HoneyHiveTracer.init(api_key=os.getenv("HH_API_KEY"))

# 3. Layer the provider instrumentor for richer model inputs/outputs
# This example uses OpenAIChatCompletion, so it uses OpenAIInstrumentor.
instrumentor = OpenAIInstrumentor()
instrumentor.instrument(tracer_provider=tracer.provider)

# 4. Import and use Semantic Kernel
from semantic_kernel.agents import ChatCompletionAgent
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion

What Gets Traced

This integration uses dual-layer tracing. Semantic Kernel’s native OpenTelemetry diagnostics emit orchestration spans, while the provider instrumentor (e.g. OpenAIInstrumentor) adds detailed model payloads. From SK-native diagnostics:

Agent invocations - ChatCompletionAgent calls with agent name and instructions
Function / plugin calls - Each kernel function execution with arguments and results
Multi-agent orchestration - Agents-as-plugins routing with triage and delegation structure

From the provider instrumentor:

LLM calls with full payloads - Chat messages, completions, tool calls, and token usage (requires SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE)

No manual instrumentation required.

Example: Agent with Plugin

import asyncio
import os
from typing import Annotated

# Enable SK OTel before any imports
os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS"] = "true"
os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE"] = "true"

from honeyhive import HoneyHiveTracer
from openinference.instrumentation.openai import OpenAIInstrumentor

tracer = HoneyHiveTracer.init(api_key=os.getenv("HH_API_KEY"))

instrumentor = OpenAIInstrumentor()
instrumentor.instrument(tracer_provider=tracer.provider)

from semantic_kernel.agents import ChatCompletionAgent
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion
from semantic_kernel.functions import kernel_function


class MathPlugin:
    @kernel_function(description="Evaluate a math expression")
    def calculate(
        self, expression: Annotated[str, "The math expression"]
    ) -> str:
        return str(eval(expression))


agent = ChatCompletionAgent(
    service=OpenAIChatCompletion(ai_model_id="gpt-4o-mini"),
    name="MathAssistant",
    instructions="Use the calculate function to solve math problems.",
    plugins=[MathPlugin()],
)


async def main():
    response = await agent.get_response(messages="What is 25 * 4?")
    print(response.content)


asyncio.run(main())

Example: Multi-Agent (Agents-as-Plugins)

Semantic Kernel supports using agents as plugins for orchestration patterns:

import asyncio
import os

os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS"] = "true"
os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE"] = "true"

from honeyhive import HoneyHiveTracer
from openinference.instrumentation.openai import OpenAIInstrumentor

tracer = HoneyHiveTracer.init(api_key=os.getenv("HH_API_KEY"))

instrumentor = OpenAIInstrumentor()
instrumentor.instrument(tracer_provider=tracer.provider)

from semantic_kernel.agents import ChatCompletionAgent
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion

# Specialist agents
billing_agent = ChatCompletionAgent(
    service=OpenAIChatCompletion(ai_model_id="gpt-4o-mini"),
    name="BillingAgent",
    instructions="You handle billing issues like charges and payment failures.",
)

refund_agent = ChatCompletionAgent(
    service=OpenAIChatCompletion(ai_model_id="gpt-4o-mini"),
    name="RefundAgent",
    instructions="Assist users with refund inquiries and status updates.",
)

# Triage agent with specialists as plugins
triage_agent = ChatCompletionAgent(
    service=OpenAIChatCompletion(ai_model_id="gpt-4o-mini"),
    name="TriageAgent",
    instructions="""Route requests to the appropriate specialist:
- Billing issues → BillingAgent
- Refund requests → RefundAgent""",
    plugins=[billing_agent, refund_agent],
)


async def main():
    response = await triage_agent.get_response(
        messages="I was charged twice for my subscription."
    )
    print(response.content)


asyncio.run(main())

Troubleshooting

Traces not appearing

Enable SK diagnostics first - Environment variables must be set before any imports:

import os

# ✅ Correct - set env vars first
os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS"] = "true"
os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE"] = "true"

from honeyhive import HoneyHiveTracer
tracer = HoneyHiveTracer.init(api_key=os.getenv("HH_API_KEY"))

from semantic_kernel.agents import ChatCompletionAgent

# ❌ Wrong - env vars set after imports
from semantic_kernel.agents import ChatCompletionAgent
os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS"] = "true"

Check environment variables - Ensure HH_API_KEY is set
Verify OpenAI credentials - Ensure OPENAI_API_KEY is configured
Instrument your provider - Use the matching provider instrumentor if you need model events with full inputs and outputs in HoneyHive. This page uses OpenAIInstrumentor because the examples use OpenAIChatCompletion.

Missing prompts/completions in traces

Enable sensitive diagnostics and the matching provider instrumentor to capture full message content on model events. This OpenAI example uses OpenAIInstrumentor:

os.environ["SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE"] = "true"

from openinference.instrumentation.openai import OpenAIInstrumentor

instrumentor = OpenAIInstrumentor()
instrumentor.instrument(tracer_provider=tracer.provider)

Enrich Your Traces

Add user IDs and custom metadata to Semantic Kernel traces

Custom Spans

Create spans for business logic around agent calls

Distributed Tracing

Trace agents across service boundaries

Query Trace Data

Export traces programmatically

Agent Frameworks & Platforms

Model Providers

AI Gateways

Coding Agents

Evaluation

How to trace Semantic Kernel with HoneyHive

Quick Start

What Gets Traced

Example: Agent with Plugin

Example: Multi-Agent (Agents-as-Plugins)

Troubleshooting

Traces not appearing

Missing prompts/completions in traces

Enrich Your Traces

Custom Spans

Distributed Tracing

Query Trace Data

Resources

​Quick Start

​What Gets Traced

​Example: Agent with Plugin

​Example: Multi-Agent (Agents-as-Plugins)

​Troubleshooting

​Traces not appearing

​Missing prompts/completions in traces

​Related

Enrich Your Traces

Custom Spans

Distributed Tracing

Query Trace Data

​Resources

Quick Start

What Gets Traced

Example: Agent with Plugin

Example: Multi-Agent (Agents-as-Plugins)

Troubleshooting

Traces not appearing

Missing prompts/completions in traces

Related

Resources