Langfuse

hybridFree (self-hosted)Free cloud (50k observations)$59/mo ProCustom Enterpriseopen source

Best for

Open-source LLM tracing, prompt management, and evaluation — self-hostable with broad framework support

Limitations

Smaller ecosystem than Datadog; self-hosted requires Postgres + ClickHouse; evaluation features are still maturing

Features

Llm Tracing
Trace LLM calls, tool invocations, and agent reasoning steps end-to-end
Cost Tracking
Track token usage and cost per request, per agent run, and per model
Evaluation
Score agent outputs against test datasets with automated evaluators
Prompt Management
Version, manage, and A/B test prompts in production
Real Time Monitoring
Live dashboards and alerting for agent performance metrics

Frameworks

langchainllamaindexvercel-aiopenai-agents

SDK Languages

pythonjavascripttypescript

Compliance

soc2gdpr

Langfuse

Langfuse is an open-source observability platform built specifically for LLM applications. It provides tracing, prompt management, evaluation, and dataset curation — all focused on the AI agent development lifecycle.

Its decorator-based Python SDK and JS integration make it straightforward to add tracing to LangChain, LlamaIndex, Vercel AI, and OpenAI agent workflows. The self-hosted option (MIT license) is a major differentiator for teams with data residency requirements.

Last verified: 2026-04-28Verified by: editorial