Strands Agents

Strands Agents (the strands-agents package, by AWS) has native OpenTelemetry tracing. It emits spans through the OTel global tracer provider, so once logfire.configure() has set Logfire as the global provider, Strands traces flow to Logfire automatically — no extra exporter needed.

Installation

Terminal

pip install logfire strands-agents strands-agents-tools

Usage

Set Strands’ semantic-convention options in your terminal before starting the application. The first option selects the latest generative AI attributes; the second records messages as attributes on their spans instead of separate span events:

Terminal

export OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental,gen_ai_span_attributes_only

Then call logfire.configure() before you construct your Agent:

from strands import Agent, tool

import logfire

# Logfire registers itself as the global OTel tracer provider.
logfire.configure()

@tool
def weather(city: str) -> str:
    """Get the current weather for a city."""
    return f"It's sunny in {city}."


agent = Agent(
    tools=[weather],
    # trace_attributes are attached to every span this agent produces.
    trace_attributes={'session.id': 'demo-1', 'user.id': 'you@example.com'},
)

result = agent("What's the weather in Lisbon?")
print(result)

You’ll see a trace in Logfire with the agent invocation, the model (LLM) call, and the weather tool call as a nested timeline. The model and agent spans contain standard gen_ai.input.messages and gen_ai.output.messages attributes for the conversation. Strands runs also appear in the specialized Agents view; the support matrix shows which columns each view populates.

Message content is sensitive

By default, Strands records message and system-prompt content, which can include personally identifiable information (PII) and is sent to Logfire. Use scrubbing to redact sensitive values before they leave your machine.

Caution

Do not call StrandsTelemetry().setup_otlp_exporter() when using Logfire. Per the Strands docs, you should skip StrandsTelemetry when a global tracer provider is already configured — your existing OpenTelemetry setup is used automatically. Calling setup_otlp_exporter() would register a competing provider and send traces to a separate endpoint instead of Logfire.

Model provider

A bare Agent(...) defaults to Amazon Bedrock (and needs AWS credentials). To run without AWS, pass an OpenAI model instance instead. Install it with pip install 'strands-agents[openai]', then construct OpenAIModel(model_id='gpt-4o') from strands.models.openai and pass that object as Agent(model=model). The OpenAI client reads OPENAI_API_KEY from your environment.

Managed prompts

Keep your agents’ system prompts in Prompt Management and fetch them at runtime:

Terminal

pip install 'logfire[variables]'

from strands import Agent
from pydantic import BaseModel

import logfire

logfire.configure()


class SystemInputs(BaseModel):
    role: str


system_var = logfire.template_var(
    name='prompt__strands_system',
    type=str,
    default='You are a helpful assistant.',
    inputs_type=SystemInputs,
)

with system_var.get(SystemInputs(role='a travel assistant'), label='production') as resolved:
    system_prompt = resolved.value

agent = Agent(system_prompt=system_prompt)

See Use Prompts in Your Application for the full workflow.