OpenTelemetry Qwen-Agent Instrumentation

OpenTelemetry instrumentation for the Qwen-Agent framework.

Installation

pip install opentelemetry-distro opentelemetry-exporter-otlp
opentelemetry-bootstrap -a install

pip install "qwen-agent >= 0.0.20"

pip install ./util/opentelemetry-util-genai

# Install this instrumentation (from the LoongSuite repo)
pip install ./instrumentation-loongsuite/loongsuite-instrumentation-qwen-agent

Published package name:

pip install loongsuite-instrumentation-qwen-agent

Usage

Auto-instrumentation

With loongsuite-instrumentation-qwen-agent installed, the opentelemetry_instrumentor entry point qwen_agent is registered for use with the OpenTelemetry distro.

opentelemetry-instrument \
    --traces_exporter console \
    python your_qwen_agent_app.py

Manual instrumentation

from opentelemetry.instrumentation.qwen_agent import QwenAgentInstrumentor
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor
from qwen_agent.agents import Assistant

provider = TracerProvider()
provider.add_span_processor(SimpleSpanProcessor(ConsoleSpanExporter()))

QwenAgentInstrumentor().instrument(tracer_provider=provider)

bot = Assistant(
    llm={"model": "qwen-max", "model_type": "qwen_dashscope"},
    name="my-assistant",
)
for _ in bot.run([{"role": "user", "content": "Hello!"}]):
    pass

QwenAgentInstrumentor().uninstrument()

Configuration

Export to an OTLP backend

export OTEL_SERVICE_NAME=my-qwen-agent-app
export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=<trace_endpoint>
# Optional: metrics / logs if you configure exporters globally
export OTEL_EXPORTER_OTLP_METRICS_ENDPOINT=<metrics_endpoint>

opentelemetry-instrument python your_app.py

GenAI semantic conventions and content capture

# Enable experimental GenAI semantic conventions (recommended for this instrumentation)
export OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental

# Message content capture (same env vars as other GenAI instrumentations in this repo)
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=SPAN_ONLY
# EVENT_ONLY | SPAN_AND_EVENT | NO_CONTENT

Supported components

Agent.run_nonstream() is not wrapped separately; it calls run() internally, so you still get a single invoke_agent span per run.

Model backends (inferred from model_type / class name): DashScope, OpenAI-compatible APIs, Azure OpenAI, and other Qwen-Agent–supported backends.

Visualization

Export telemetry to:

• Alibaba Cloud ARMS / Managed Service for OpenTelemetry
• AgentScope Studio or any OTLP-compatible collector
• Any OpenTelemetry-compatible backend (Jaeger, Zipkin, etc.)

License

Apache License 2.0