#11941·langfuse

Graph visualisation incorrect for AutoGen multi-agent workflows via OTLP

Author: SajidAlamQBCreated Feb 9, 2026Updated Sep 17, 2026
Labelsbug

Describe the bug

When tracing AutoGen multi-agent workflows via OpenTelemetry, OTLP, export to Langfuse, the graph visualisation renders incorrectly despite the trace data having the correct hierarchical structure.

The trace JSON shows proper parent-child relationships with correct depth levels, but the graph view displays nodes with unexpected connections or layout.

Image

Expected behaviour

Graph visualisation should reflect the actual parent-child span relationships present in the trace data.

Actual behaviour

Graph view renders with incorrect connections between nodes. The trace timeline view shows correct nesting, but the graph diagram does not match.

Environment

Langfuse Cloud
Python SDK: OpenTelemetry OTLP exporter
Framework: AutoGen with multi-agent orchestration (2+ agents with tool calls)

Related issues

This appears related to:

LFE-6958 — "The langfuse graph is not plotting correctly"
LFE-7861 — "LangGraph CallbackHandler Observations Do Not Nest Under Active Parent Span"
LFE-7159 — "Langgraph-Observations all get bundled in one trace"
LFE-5967 — "LangGraph call doesn't show sub graph nodes with the Graph Diagram"

Steps to reproduce

Configure OTLP export to Langfuse:

pythonfrom opentelemetry import trace
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor

exporter = OTLPSpanExporter(
endpoint="https://cloud.langfuse.com/api/public/otel/v1/traces",
headers={"Authorization": "Basic <base64_credentials>"}
)
provider = TracerProvider()
provider.add_span_processor(BatchSpanProcessor(exporter))
trace.set_tracer_provider(provider)
tracer = trace.get_tracer("langfuse.autogen")

Wrap AutoGen agent execution with a parent span:

pythonwith tracer.start_as_current_span("response_generation") as span:
span.set_attribute("intent", "claim_new")
# AutoGen multi-agent orchestration
agent.invoke(context)

View the trace in Langfuse the graph view shows incorrect node connections/layout, even though the trace timeline and JSON show correct hierarchy.

Langfuse Cloud or self-hosted?

Langfuse Cloud

If self-hosted, what version are you running?

No response

SDK and integration versions

No response

Additional information

No response

Are you interested in contributing a fix for this bug?

No