Langfuse: LLM Observability Platform

Overview

Langfuse is an open-source LLM engineering platform covering observability, prompt management, evaluations, and cost tracking in a single product. It integrates with LangChain, LiteLLM, the OpenAI SDK, OpenTelemetry, and most major agent frameworks. The MIT-licensed core can be self-hosted; Langfuse Cloud is the managed option.

Pricing (Cloud)

Plan	Price	Units/month
Hobby	Free	50,000
Core	$29/month	More units + longer retention
Pro	$199/month	Team features, advanced evals
Enterprise	$2,499/month	SSO, SAML, data residency, SLAs

Overages across all paid plans: $8 per 100,000 additional units. A "unit" covers one observed event (an LLM call, a retrieval step, a tool execution). Self-hosted has no unit limits or licensing fees.

What it covers

Each trace captures every operation in an LLM call chain: inputs, outputs, timing, model version, token counts, and cost. The prompt management feature lets you version, deploy, and roll back prompts independently of application deploys -- a one-click rollback to a prompt version is faster than a code deploy.

Evaluations include: LLM-as-a-judge scoring, user feedback collection, manual labeling, and custom pipelines via the API. This connects directly to the Datasets feature for running eval suites against specific prompt versions.

Where it fits

Langfuse is the observability layer for teams that want full trace visibility into agent execution and want to iterate on prompts without code deploys. The free tier (50,000 units/month) covers most small production applications.

Compared to Helicone: Langfuse is deeper on eval and prompt management; Helicone is simpler to instrument (one-line proxy change) and covers the gateway layer. Teams that need both observability depth and gateway features sometimes run both.

For teams using LangChain, the native LangSmith observability is a direct alternative. Langfuse is the choice when you want open-source with self-host optionality.

Field notes

• Langfuse reached 10,000 GitHub stars by April 2026 per its GitHub repository. The growth correlates with LangChain adoption -- teams instrumenting LangChain pipelines reach for Langfuse as the open-source LangSmith alternative. [community-thread, 2026-04-05]

See also

Helicone, Portkey, LangChain