Langfuse

Langfuse is an open-source LLM engineering and observability platform that connects tracing, prompt management, evaluation, and experimentation into a continuous loop for AI applications and agents. Development teams use it to monitor production data, analyze latency and costs, run evaluations, and collaborate on fixes from prototype to production scale. The platform can be deployed via Langfuse Cloud or self-hosted using Docker Compose, Kubernetes, or Terraform configurations for AWS, GCP, and Azure.

• Hierarchical traces capture LLM calls, tool invocations, and retrieval steps with filters for user, session, cost, latency, or custom metadata.
• Evaluation capabilities include LLM-as-a-judge, heuristic functions, human review, and regression datasets.
• Prompt management separates prompts from code with one-click deployments and rollbacks.
• Playground enables testing prompts on real production inputs and comparing models side-by-side.
• Experiments define test cases and run comparisons side-by-side.
• Human annotation supports collaborative human-in-the-loop workflows to review traces and create golden datasets.
• Dashboards and automated alerts monitor cost, token spend, latency, and quality.
• Langfuse Assistant and developer tooling such as SKILL.md, a CLI, and a platform MCP server connect coding agents to the platform.

Pros

• Provides complete data portability under an MIT license without vendor lock-in.
• Supports flexible self-hosting deployment options including Docker Compose, Kubernetes, and Terraform scripts for major cloud providers.
• Integrates natively with multiple languages and frameworks via OpenTelemetry and native SDKs without framework lock-in.
• Combines tracing, prompt management, evaluation, and experimentation in a single integrated workflow.

Cons

• Cloud hosting plans enforce unit limits per month, requiring additional fees for overages beyond the included allocations.
• Lower-tier plans have restricted data access windows, such as 30 days on the Hobby tier and 90 days on the Core tier.
• Advanced security features like audit logs, SCIM APIs, and SSO require higher-tier plans or the Teams add-on.
• Dedicated support channels and custom SLAs are restricted to enterprise-level tiers.

This platform fits development teams and AI engineers building applications and agents who require deep visibility into production LLM traffic, prompt management, and evaluation loops. It suits teams wanting an open-source tool with flexible self-hosting or managed cloud options. It is a poor fit for organizations that do not utilize LLMs or AI agents, or those seeking legacy observability tools that handle only small spans without verbose I/O payloads.

Langfuse offers a freemium model with options for both Langfuse Cloud hosting and self-hosting. The Hobby plan is free, includes 50,000 units per month, 30 days of data access, unlimited users, and community support. The Core plan costs $29 per month, includes 100,000 units per month, 90 days of data access, unlimited users, and in-app support. The Pro plan costs $199 per month, includes 100,000 units per month, 3 years of data access, data retention management, high rate limits, SOC2 and ISO27001 reports, and prioritized support. An optional Teams Add-on is available for $300 per month to add enterprise SSO and fine-grained RBAC. The Enterprise plan costs $2,499 per month and adds audit logs, SCIM APIs, uptime SLAs, and a dedicated support engineer. Additional units across paid plans cost $8 per 100k units, with volume discounts available.

Unlike traditional observability platforms built for legacy application spans, Langfuse is engineered specifically to handle the rich, verbose I/O generated by LLM systems at scale. It distinguishes itself from proprietary alternatives by being fully open source under an MIT license with guaranteed data portability and no vendor lock-in. Rather than functioning solely as a monitoring dashboard, it combines tracing, evaluations, experiments, and prompt management into a unified loop that integrates directly with coding agents and terminal workflows.

Langfuse works with Python, TypeScript, Go, Java, .NET, Ruby, PHP, and Swift via OpenTelemetry and native SDKs. Agent frameworks include LangChain, Vercel AI SDK, LiteLLM, Pydantic AI, Google ADK, CrewAI, LiveKit, LangChain DeepAgents, OpenWebUI, Ollama, OpenAI Agents SDK, Dify, Langflow, OpenRouter, n8n, Spring AI, Cursor, PostHog, DSPy, Amazon AgentCore, Strands Agents, LlamaIndex, Agno Agents, Temporal, ClickHouse Agentic Data Stack, Mastra, Claude Agent SDK for JS, Promptfoo, Microsoft Agent Framework, Google Vertex AI, Ragas, AutoGen, and RAGflow. Model providers include OpenAI, Anthropic, Amazon Bedrock, Azure OpenAI, Mistral AI, Google Gemini, xAI, vLLM, and Groq.