OTel / OTLP

MeshFlow Cloud exposes an OTLP-compatible endpoint. Any OpenTelemetry-instrumented agent can send spans directly — no MeshFlow SDK required. Spans are converted to ATR records and appear in the Trace Explorer immediately.

GenAI semantic conventions

MeshFlow maps gen_ai.* OTel semantic conventions to ATR fields automatically: gen_ai.system → modelVersion, gen_ai.prompt → inputText.

OTLP endpoint

bash

# Set your OTLP exporter endpoint to MeshFlow Cloud
export OTEL_EXPORTER_OTLP_ENDPOINT=https://ingest.meshflow.ai/otlp
export OTEL_EXPORTER_OTLP_HEADERS="x-meshflow-key=mf-..."

Auto-instrument

Use any OpenTelemetry auto-instrumentation library. MeshFlow accepts standard OTLP/HTTP and OTLP/gRPC.

python

from opentelemetry.sdk.trace import TracerProvider

from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter

from opentelemetry.sdk.trace.export import BatchSpanProcessor

exporter = OTLPSpanExporter(

endpoint="https://ingest.meshflow.ai/otlp/v1/traces",

headers={"x-meshflow-key": os.getenv("MESHFLOW_API_KEY")},

)

provider = TracerProvider()

provider.add_span_processor(BatchSpanProcessor(exporter))

Semantic conventions

MeshFlow maps the following gen_ai.* attributes to ATR fields when ingesting OTel spans:

OTel attribute	ATR field
`gen_ai.system`	modelVersion
`gen_ai.request.model`	modelVersion
`gen_ai.prompt`	inputText
`gen_ai.completion`	outputText
`gen_ai.usage.prompt_tokens`	inputTokens
`gen_ai.usage.completion_tokens`	outputTokens

TypeScript SDK

AutoGen 0.4+

Docs/Integrations/OTel / OTLP