Langfuse

Langfuse is an open source LLM engineering platform built by Langfuse GmbH (now part of ClickHouse) that helps teams develop, monitor, evaluate, and debug AI applications and agents. It is MIT-licensed for core features, self-hostable in minutes, and available as a managed cloud service. The platform is built on OpenTelemetry and supports over 80 integrations across model providers, agent frameworks, and languages.

What It Is

Langfuse sits in the LLMOps category, providing the full development lifecycle toolchain for teams building production-grade LLM applications. It combines LLM observability (tracing), prompt management, evaluation pipelines, a playground, and experiment datasets into one integrated platform. Engineers can use it standalone for tracing or adopt the full suite to power a continuous improvement loop from prototype to production.

Core Platform Capabilities

• LLM Observability: Hierarchical traces capture every LLM call, tool invocation, retrieval step, and agent action. Traces include session tracking, user tracking, token and cost tracking, and agent graph visualization.
• Prompt Management: Version-controlled prompts with one-click deployments and rollbacks, server- and client-side caching, composability, and release labels — all without code changes.
• Evaluation: LLM-as-a-judge, user feedback, manual annotation queues, and custom evaluation pipelines via API/SDK. Evaluations run on production traces or against offline datasets.
• Experiments & Datasets: Define test cases, run structured experiments, and compare results side-by-side in the UI or via SDK.
• Playground: Test prompts on real production inputs and compare models side-by-side directly from a trace.
• Metrics & Dashboards: Monitor cost, latency, and quality with dashboards and automated alerts.

Integration Breadth and Stack Compatibility

Langfuse is built on OpenTelemetry, which means it works with any language that supports OTel instrumentation (Python, TypeScript, Go, Java, .NET, Ruby, PHP, Swift). Native SDKs exist for Python and JavaScript/TypeScript. The platform lists 80+ integrations including:

• Agent frameworks: LangChain, LlamaIndex, Vercel AI SDK, CrewAI, Pydantic AI, Google ADK, OpenAI Agents SDK, Mastra, AutoGen, DSPy, and more
• Model providers: OpenAI, Anthropic, Amazon Bedrock, Azure OpenAI, Mistral AI, Google Gemini, xAI, vLLM, Groq, Ollama
• No-code tools: Dify, Langflow, OpenWebUI, n8n
• Analytics: PostHog, Mixpanel

Deployment Model

Langfuse offers two deployment paths: a managed cloud (Langfuse Cloud) with US, EU, and JP data regions, and a fully self-hosted option. Self-hosting is supported via Docker Compose, Kubernetes (Helm), and Terraform templates for AWS, GCP, and Azure. The core platform is MIT-licensed and all product features are available in the self-hosted version. The architecture uses a ClickHouse OLAP database, async ingestion via Redis queue, and S3/blob storage for large payloads — designed to handle billions of observations per month.

Update: Joining ClickHouse and Recent Activity

In January 2026, Langfuse joined ClickHouse, the open-source database company, to accelerate development. The latest OSS release as of the source data is v3.174.1 (published May 13, 2026), with the repository showing active daily releases. Recent changelog entries include "Sign in with ClickHouse Cloud" (May 18, 2026), "Introducing Langfuse Academy" (May 14, 2026), and "Self-Service Enterprise SSO Setup" (May 8, 2026). The homepage states the project has over 27,400 GitHub stars and 300+ contributors.

Enterprise Scale and Security

The platform is designed for high-volume LLM workloads. According to the Langfuse website, it processes over 10 billion observations per month and has over 50 million SDK installs per month. Security certifications listed on the site include SOC 2 Type II, ISO 27001, GDPR compliance, and HIPAA eligibility. Enterprise deployments support SCIM, audit logs, fine-grained RBAC, enterprise SSO (Okta, AzureAD/EntraID), custom rate limits, and uptime SLAs.