Langfuse

Overview

Langfuse is an open-source LLM engineering platform that helps teams collaboratively debug, analyze, and iterate on their LLM applications. Provides LLM Observability, metrics, evals, prompt management, playground, and datasets.

Core Features

LLM Observability

• Purpose-built for LLM applications with native understanding of LLM-specific concepts
• Captures token usage, model parameters, prompt/completion pairs, and evaluation scores
• Application tracing for every operation: LLM calls, retrieval steps, tool executions, and custom logic
• Timing, inputs, outputs, and metadata tracking

Prompt Management

• Systematic approach to storing, versioning, and retrieving prompts
• Central management instead of hardcoding prompts in application code
• Version control for prompt iterations

Vector Database Integration

• Monitors RAG pipelines that fetch relevant context from vector databases (ChromaDB, etc.)
• Tracks document retrieval operations
• Vector search monitoring (VectorStoreRetriever)
• Answer generation tracking

OpenTelemetry Support

OTEL-native Langfuse SDK v3 is a thin layer on top of official OpenTelemetry client that:

• Automatically converts spans into rich Langfuse observations
• Provides first-class helpers for token usage and cost tracking
• Supports prompt linking and scoring

Integrations

• OpenTelemetry
• LangChain
• OpenAI SDK
• LiteLLM
• ChromaDB and other vector databases