Haystack

Haystack is deepset’s framework for building agents and retrieval pipelines (package haystack-ai). You can send traces of agent steps, tool calls, model requests, and pipeline components to Logfire.

Haystack’s first-party opentelemetry-haystack package connects its tracing system to OpenTelemetry, the open standard Logfire uses to receive telemetry. logfire.configure() installs the application-wide OpenTelemetry destination, and OpenTelemetryTracer sends Haystack’s own spans to it.

Installation

Terminal

pip install logfire "haystack-ai>=3.0" "opentelemetry-haystack>=1.0"

Usage

Haystack omits prompt, response, tool argument, and tool result content from its spans by default. Set HAYSTACK_CONTENT_TRACING_ENABLED=true before importing Haystack to capture that content. This sends potentially sensitive application data to Logfire, so leave it disabled if that data should not leave your application.

import os

os.environ['HAYSTACK_CONTENT_TRACING_ENABLED'] = 'true'

from haystack import tracing
from haystack.components.agents import Agent
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.dataclasses import ChatMessage
from haystack.tools import Tool
from haystack_integrations.tracing.opentelemetry import OpenTelemetryTracer
from opentelemetry import trace

import logfire

logfire.configure()
tracing.enable_tracing(OpenTelemetryTracer(trace.get_tracer('haystack')))


def lookup_incident(incident_id: str) -> str:
    return f'{incident_id} is resolved; owner=platform-observability'


lookup_tool = Tool(
    name='lookup_incident',
    description='Look up the current status and owner of an incident by ID.',
    parameters={
        'type': 'object',
        'properties': {'incident_id': {'type': 'string'}},
        'required': ['incident_id'],
    },
    function=lookup_incident,
)

agent = Agent(
    chat_generator=OpenAIChatGenerator(model='gpt-4o-mini'),  # needs OPENAI_API_KEY
    tools=[lookup_tool],
    system_prompt='Use operational tools to verify facts before answering.',
    max_agent_steps=3,
)
result = agent.run(
    messages=[ChatMessage.from_user("Use lookup_incident for incident_id='incident-42'.")]
)
print(result['last_message'].text)

You’ll see Haystack’s haystack.agent.run trace in Logfire, with child spans for each agent step, model request, and tool call.

Caution

Don’t also enable openinference-instrumentation-haystack for the same application. Both integrations trace Haystack operations, so enabling both can produce duplicate spans.

Managed prompts

Keep your pipeline’s prompt templates in Prompt Management and fetch them at runtime:

Terminal

pip install 'logfire[variables]'

from haystack.dataclasses import ChatMessage
from pydantic import BaseModel

import logfire

logfire.configure()


class FactInputs(BaseModel):
    topic: str


prompt_var = logfire.template_var(
    name='prompt__fun_fact',
    type=str,
    default='Tell me a one-line fun fact about {{topic}}.',
    inputs_type=FactInputs,
)

with prompt_var.get(FactInputs(topic='the Roman Empire'), label='production') as resolved:
    user_message = ChatMessage.from_user(resolved.value)

# Pass `user_message` straight to your generator / pipeline.

See Use Prompts in Your Application for the full workflow.