Langfuse Python SDK
[!IMPORTANT]
The SDK was rewritten in v4 and released in March 2026. Refer to the v4 migration guide for instructions on updating your code.
This is the latest Langfuse Python SDK (v4, OpenTelemetry). The legacy v2 client API (trace(), span(), generation(), api.trace.list) is deprecated. Do not use it for new instrumentation.
Latest Python SDK (OpenTelemetry): pip install langfuse. Use get_client() + start_as_current_observation / @observe. Prompts/datasets/scores: the same langfuse package. Docs: https://langfuse.com/docs/observability/sdk/overview. Upgrade: v2 to v3, v3 to v4.
Query traces with Observations API v2 (GET /api/public/v2/observations, API reference), not GET /api/public/traces / api.trace.list.
Data freshness: the only real-time path is OpenTelemetry ingestion (this SDK at langfuse >= 4.7.0, or any OTel exporter sending x-langfuse-ingestion-version: 4) combined with the Observations API v2 and Metrics API v2. All other GET APIs from all SDK versions can be up to 10 minutes behind.
The Langfuse Python SDK covers the full platform: observability/tracing (OpenTelemetry-based, with OpenAI and LangChain integrations), datasets & experiments (offline evaluation and regression testing of prompt/model changes, including CI via GitHub Actions), LLM-as-a-judge and custom evaluations/scores, prompt management, and a full REST API client.
Installation
pip install langfuse
Quickstart
# env: LANGFUSE_PUBLIC_KEY, LANGFUSE_SECRET_KEY, LANGFUSE_BASE_URL
from langfuse import get_client
langfuse = get_client()
# Create a span using a context manager
with langfuse.start_as_current_observation(as_type="span", name="process-request") as span:
# Your processing logic here
span.update(output="Processing complete")
# Create a nested generation for an LLM call
with langfuse.start_as_current_observation(as_type="generation", name="llm-response", model="gpt-5.6") as generation:
# Your LLM call logic here
generation.update(output="Generated response")
# All spans are automatically closed when exiting their context blocks
# Flush events in short-lived applications
langfuse.flush()
Docs
