The SDK includes an optional OpenTelemetry exporter that emits OTel spans for traces captured by Infinium. This lets you bridge Infinium traces into your existing OTel pipeline.

Requirements

Install the OpenTelemetry dependencies (not included with the SDK):

pip install opentelemetry-api opentelemetry-sdk

Setup

from infinium.integrations.otel import InfiniumOTelExporter

exporter = InfiniumOTelExporter(
    service_name="my-agent",    # Service identifier for spans
    tracer_name="infinium",     # Tracer name (default: "infinium")
)

Parameters

Parameter	Type	Default	Description
service_name	str	"infinium-agent"	Service name for OTel resource
tracer_name	str	"infinium"	Tracer name

Exporting a Trace

Call export_trace() to create OTel spans from trace data:

from infinium.integrations._context import TraceContext

exporter.export_trace(
    trace_name="Email Classifier",
    duration_s=2.4,
    ctx=trace_context,  # Optional TraceContext with captured calls
)

Parameters

Parameter	Type	Default	Description
trace_name	str	required	Name for the root span
duration_s	float	required	Trace duration in seconds
ctx	TraceContext	None	Context with captured LLM calls and prompt fetches

Span Attributes

Root Span

Attribute	Description
infinium.service	Service name
infinium.duration_s	Total duration in seconds
infinium.llm_call_count	Number of LLM calls in the trace
infinium.prompt_fetch_count	Number of prompt fetches in the trace

LLM Call Child Spans

Each captured LLM call is emitted as a child span:

Attribute	Description
llm.provider	Provider name (openai, anthropic, google, xai)
llm.model	Model identifier
llm.prompt_tokens	Input token count
llm.completion_tokens	Output token count
llm.temperature	Temperature parameter
llm.latency_ms	Call latency in milliseconds
error.type	Error type (if the call failed)

Prompt Fetch Events

Prompt fetches are recorded as span events (not child spans):

Attribute	Description
prompt.id	Prompt UUID
prompt.name	Prompt display name
prompt.version	Version number

Example: Full Pipeline

from openai import OpenAI
from infinium import InfiniumClient
from infinium.integrations import watch
from infinium.integrations.otel import InfiniumOTelExporter
from infinium.integrations._context import get_current_trace_context

# Set up OTel exporter
exporter = InfiniumOTelExporter(service_name="research-agent")

# Set up Infinium client with auto-instrumentation
client = InfiniumClient(agent_id="...", agent_secret="...")
openai = watch(OpenAI())

@client.trace("Research Task")
def research(query: str) -> str:
    resp = openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": query}],
    )
    return resp.choices[0].message.content

result = research("Latest AI trends")

# Export the same trace data to OTel
# (In practice, you'd integrate this into your trace pipeline)

Notes

• The OTel exporter is optional — it does not affect normal SDK operation
• If opentelemetry-api or opentelemetry-sdk are not installed, importing the exporter will raise an ImportError
• The exporter creates spans using the standard OTel API, so they integrate with any configured OTel pipeline (Jaeger, Zipkin, OTLP, etc.)