Langfuse overview
Category: Model APIs and platforms
Langfuse is an open-source platform for tracing, evaluating, and managing LLM and agent applications, with an optional AI Gateway. Official pages used here are langfuse.com and langfuse.com/pricing, captured 2026-11-11 (Asia/Colombo). This overview is desk research, not an observability trial, and on AI Stack Wise this is a directory overview. Nothing on this page reflects personal hands-on testing.
What the official product is
The official site describes one platform for every AI call a company makes: usage and cost tracking, traces and evaluations, prompt and skill management, datasets and experiments, and production logs that can be exported to a warehouse. You can instrument with official SDKs or OpenTelemetry, or route OpenAI and Anthropic calls through the Langfuse AI Gateway by swapping the base URL and API key so every call is traced automatically, with cost shown per user and key. You can also bring your own gateway. An in-app agent named Halo, or your own agent, can work with this data, and features are exposed via MCP, CLI, API, and SDKs.
The intended loop is explicit: in production, trace and monitor; in development, build datasets, experiment, and evaluate; then deploy and start again. Evaluators can be LLM-as-a-judge, heuristic functions, or human review, and can run on production data or during experiments. Prompts can be separated from code and deployed independently. Langfuse Academy is a free, open set of materials on that loop, also readable as Markdown for coding agents.
Official scale claims on the pricing FAQ include 50,000+ companies, billions of events per month, and 65M+ SDK installs per month. Those are vendor figures, not our telemetry.
You can use Langfuse Cloud in global regions, or self-host with Docker Compose, Kubernetes (Helm), or Terraform templates for AWS, GCP, and Azure. Self-hosting is described as free under the MIT license, with core features — observability, evaluation, prompt management, datasets, and the AI Gateway — included at unlimited usage; you pay only for your own infrastructure. A self-hosted Enterprise plan adds project-level RBAC, audit logs, data-retention policies, management APIs, and dedicated support with SLAs.
Official limits and how it is sold
A recapture of the official pricing page on 2026-11-12 (Asia/Colombo) prints a clear Cloud plan table in USD. Hobby is free with no credit card, 50k units a month, 30 days of data access, and 2 users. Core is $29 a month, Pro $199 a month, and Enterprise $2,499 a month. Each paid plan includes 100k units a month, then graduated usage: $8 per 100k units from 100k to 1M, $7 from 1M to 10M, $6.50 from 10M to 50M, and $6 above 50M. An optional Teams add-on for Pro (enterprise SSO, SSO enforcement, fine-grained RBAC) is $300 a month. Billable units are the traces, observations, and scores you send. The plan fee is charged at the start of the month and usage at the end. Tax treatment is not stated. Our earlier 2026-11-11 capture had conflicting free-volume wording; the 2026-11-12 plan table resolves it at 50k units for Hobby.
Data access differs by plan (30 days, 90 days, 3 years), and the official FAQ separates it from opt-in data retention on Pro and Enterprise. Discounts are listed for early-stage startups, research and students, non-profits, hackathons, and open-source projects. Current plan prices, and any gaps we could not verify, are on our Langfuse pricing page.
Who it fits
Langfuse fits teams that want open-source tracing and evals they can self-host for infrastructure cost only, or a managed Cloud with a usage-based bill. It is a weaker fit if you only need a generic APM for non-LLM spans, or if your trace volume is large enough that graduated usage, not the plan fee, dominates the bill — model that with the official calculator.
Related: